In this paper, we present a novel approach to remove object motion blur in 3D scene renderings using deformable neural radiance fields. Our technique adapts the hyperspace representation to accommodate shape changes induced by object motion blur. Experiments on Blender-generated datasets demonstrate the effectiveness of our method in producing higher-quality images with reduced object motion blur artifacts.

We present an interactive route viewer system, 3DMovieMap, which generates and shows navigation movies walking through multi-level buildings, such as a science museum, airport, and university building. Movie map systems can provide users with visual cues by synthesizing navigation movies based on their inputs of routes. However, existing systems are limited to flat areas such as city areas. We aim to extend Movie Map to generate navigation movies for multi-level buildings. The 3DMovieMap system generates a movie map from an equirectangular movie via a visual Simultaneous Localization and Mapping technology. Users select waypoints on the floor maps. 3DMovieMap calculates the shortest path that visits these points and generates a navigation movie along the route. We created four movie maps of buildings and asked two participants to use our system and provide feedback for further improvements. We will be releasing an open dataset of equirectangular movies captured in a science museum.

Indoor navigation systems with prebuilt maps have shown great potential in navigating blind people even in unfamiliar buildings. However, blind people cannot always benefit from them in every building, as prebuilt maps are expensive to build. This paper explores a map-less navigation system for blind people to reach destinations in unfamiliar buildings, which is implemented on a robot. We first conducted a participatory design with five blind people, which revealed that intersections and signs are the most relevant information in unfamiliar buildings. Then, we prototyped PathFinder, a navigation system that allows blind people to determine their way by detecting and conveying information about intersections and signs. Through a participatory study, we improved the interface of PathFinder, such as the feedback for conveying the detection results. Finally, a study with seven blind participants validated that PathFinder could assist users in navigating unfamiliar buildings with increased confidence compared to their regular aid.

Enabling blind visitors to explore museum floors while feeling the facility's atmosphere and increasing their autonomy and enjoyment are imperative for giving them a high-quality museum experience. We designed a science museum exploration system for blind visitors using an autonomous navigation robot. Blind users can control the robot to navigate them toward desired exhibits while playing short audio descriptions along the route. They can also browse detailed explanations on their smartphones and call museum staff if interactive support is needed. Our real-world user study at a science museum during its opening hour revealed that blind participants could explore the museum safely and independently at their own pace. The study also showed that the sighted visitors who saw the participants walking with the robot accepted the assistive robot well. We finally conducted focus group sessions with the blind participants and discussed further requirements toward a more independent museum experience.

Training opportunities for visually impaired (VI) skiers are limited because it is essential for them to have sighted people who guide them with their voices. This study investigates an auditory feedback system that enables ski training using a ski simulator for VI skiers alone. Based on the results of interviews with actual VI skiers and their guides, we designed the following three types of sounds: 1) a single sound (ATS: Advance Turn Sound) that conveys information about turns; 2) a continuous sound (CES: Continuous Error Sound) that is emitted according to the difference between the user's future position and the position he/she should progress to; and 3) a single sound (Gate-Passed Sound) which is emitted when a user passed through a gate. We conducted an evaluation experiment with four blind skiers and three sighted guides. Results showed that three out of four skiers performed better under the conditions in which ATS and gate-passed sound were emitted than the condition in which a human guide gave calls. The result suggests that a sonification-based method such as ATS is effective for ski training on the ski simulator for VI skiers.

Effective communication between humans and intelligent agents has promising applications for solving complex problems. One such approach is visual dialogue, which leverages multimodal context to assist humans. However, real-world scenarios occasionally involve human mistakes, which can cause intelligent agents to fail. While most prior research assumes perfect answers from human interlocutors, we focus on a setting where the agent points out unintentional mistakes for the interlocutor to review, better reflecting real-world situations. In this paper, we show that human answer mistakes depend on question type and QA turn in the visual dialogue by analyzing a previously unused data collection of human mistakes. We demonstrate the effectiveness of those factors for the model's accuracy in a pointing-human-mistake task through experiments using a simple MLP model and a Visual Language Model.

People with visual impairment (PVI) are eager to push their limits in extreme sports such as alpine skiing. However, training skiing is very difficult as it always requires assistance from an experienced guide. This paper explores sonification-based methods that enable PVI to train skiing using a simulator, which will allow them to train without a guide. Two types of sonification feedback for PVI are proposed and studied in our experiment. The results suggest that users without any visual information can also pass through over 80% of the poles compared to with visual information on average.

This paper describes an unsupervised disentangled representation learning method for musical instrument sounds with pitched and unpitched spectra. Since conventional methods have commonly attempted to disentangle timbral features (e.g., instruments) and pitches (e.g., MIDI note numbers and FOs), they can be applied to only pitched sounds. Global timbres unique to instruments and local variations (e.g., expressions and playstyles) are also treated without distinction. Instead, we represent the spectrogram of a musical instrument sound with a variational autoencoder (VAE) that has timbral, pitch, and variation features as latent variables. The pitch clarity or percussiveness, brightness, and FOs (if existing) are considered to be represented in the abstract pitch features. The unsupervised disentanglement is achieved by extracting time-invariant and time-varying features as global timbres and local variations from randomly pitch-shifted input sounds and time-varying features as local pitch features from randomly timbre-distorted input sounds. To enhance the disentanglement of timbral and variation features from pitch features, input sounds are separated into spectral envelopes and fine structures with cepstrum analysis. The experiments showed that the proposed method can provide effective timbral and pitch features for better musical instrument classification and pitch estimation.

Navigating in an indoor corridor can be challenging for blind people as they have to be aware of obstacles while also having to recognize the intersections that lead to the destination. To aid blind people in such tasks, we propose Corridor-Walker, a smartphone-based system that assists blind people to avoid obstacles and recognize intersections. The system uses a LiDAR sensor equipped with a smartphone to construct a 2D occupancy grid map of the surrounding environment. Then, the system generates an obstacle-avoiding path and detects upcoming intersections on the grid map. Finally, the system navigates the user to trace the generated path and notifes the user of each intersection’s existence and the shape using vibration and audio feedback. A user study with 14 blind participants revealed that Corridor-Walker allowed participants to avoid obstacles, rely less on the wall to walk straight, and enable them to recognize intersections.

This paper presents an audio-to-animation synthesis method for violin performance. This new approach provides a fine-grained violin performance animation using information on playing procedure consisting of played string, finger number, position, and bow direction. We demonstrate that our method is capable of synthesizing natural violin performance animation with fine fingering and bowing through extensive evaluation.

Autonomous navigation robots have a considerable potential to offer a new form of mobility aid to people with visual impairments. However, to deploy such robots in public buildings, it is imperative to receive acceptance from not only robot users but also people that use the buildings and managers of those facilities. Therefore, we conducted three studies to investigate the acceptance and concerns of our prototype robot, which looks like a regular suitcase. First, an online survey revealed that people could accept the robot navigating blind users. Second, in the interviews with facility managers, they were cautious about the robot's camera and the privacy of their customers. Finally, focus group sessions with legally blind participants who experienced the robot navigation revealed that the robot may cause trouble when it collides with those who may not be aware of the user's blindness. Still, many participants liked the design of the robot which assimilated into the surroundings.

In the task of audio-visual sound source separation, which leverages visual
information for sound source separation, identifying objects in an image is a
crucial step prior to separating the sound source. However, existing methods
that assign sound on detected bounding boxes suffer from a problem that their
approach heavily relies on pre-trained object detectors. Specifically, when
using these existing methods, it is required to predetermine all the possible
categories of objects that can produce sound and use an object detector
applicable to all such categories. To tackle this problem, we propose a fully
unsupervised method that learns to detect objects in an image and separate
sound source simultaneously. As our method does not rely on any pre-trained
detector, our method is applicable to arbitrary categories without any
additional annotation. Furthermore, although being fully unsupervised, we found
that our method performs comparably in separation accuracy.

The present paper introduces a group activity involving writing summaries of
conference proceedings by volunteer participants. The rapid increase in
scientific papers is a heavy burden for researchers, especially non-native
speakers, who need to survey scientific literature. To alleviate this problem,
we organized a group of non-native English speakers to write summaries of
papers presented at a computer vision conference to share the knowledge of the
papers read by the group. We summarized a total of 2,000 papers presented at
the Conference on Computer Vision and Pattern Recognition, a top-tier
conference on computer vision, in 2019 and 2020. We quantitatively analyzed
participants' selection regarding which papers they read among the many
available papers. The experimental results suggest that we can summarize a wide
range of papers without asking participants to read papers unrelated to their
interests.

Single-view depth estimation from omnidirectional images has gained
popularity with its wide range of applications such as autonomous driving and
scene reconstruction. Although data-driven learning-based methods demonstrate
significant potential in this field, scarce training data and ineffective 360
estimation algorithms are still two key limitations hindering accurate
estimation across diverse domains. In this work, we first establish a
large-scale dataset with varied settings called Depth360 to tackle the training
data problem. This is achieved by exploring the use of a plenteous source of
data, 360 videos from the internet, using a test-time training method that
leverages unique information in each omnidirectional sequence. With novel
geometric and temporal constraints, our method generates consistent and
convincing depth samples to facilitate single-view estimation. We then propose
an end-to-end two-branch multi-task learning network, SegFuse, that mimics the
human eye to effectively learn from the dataset and estimate high-quality depth
maps from diverse monocular RGB images. With a peripheral branch that uses
equirectangular projection for depth estimation and a foveal branch that uses
cubemap projection for semantic segmentation, our method predicts consistent
global depth while maintaining sharp details at local regions. Experimental
results show favorable performance against the state-of-the-art methods.

3D car models are heavily used in computer games, visual effects, and even automotive designs. As a result, producing such models with minimal labour costs is increasingly more important. To tackle the challenge, we propose a novel system to reconstruct a 3D car using a single sketch image. The system learns from a synthetic database of 3D car models and their corresponding 2D contour sketches and segmentation masks, allowing effective training with minimal data collection cost. The core of the system is a machine learning pipeline that combines the use of a generative adversarial network (GAN) and lazy learning. GAN, being a deep learning method, is capable of modelling complicated data distributions, enabling the effective modelling of a large variety of cars. Its major weakness is that as a global method, modelling the fine details in the local region is challenging. Lazy learning works well to preserve local features by generating a local subspace with relevant data samples. We demonstrate that the combined use of GAN and lazy learning produces is able to produce high-quality results, in which different types of cars with complicated local features can be generated effectively with a single sketch. Our method outperforms existing ones using other machine learning structures such as the variational autoencoder.

This paper describes a deep learning-based method for long-term video interpolation that generates intermediate frames between two music performance videos of a person playing a specific instrument. Recent advances in deep learning techniques have successfully generated realistic images with high-fidelity and high-resolution in short-term video interpolation. However, there is still room for improvement in long-term video interpolation due to lack of resolution and temporal consistency of the generated video. Particularly in music performance videos, the music and human performance motion need to be synchronized. We solved these problems by using human poses and music features essential for music performance in long-term video interpolation. By closely matching human poses with music and videos, it is possible to generate intermediate frames that synchronize with the music. Specifically, we obtain the human poses of the last frame of the first video and the first frame of the second video in the performance videos to be interpolated as key poses. Then, our encoder-decoder network estimates the human poses in the intermediate frames from the obtained key poses, with the music features as the condition. In order to construct an end-to-end network, we utilize a differentiable network that transforms the estimated human poses in vector form into the human pose in image form, such as human stick figures. Finally, a video-to-video synthesis network uses the stick figures to generate intermediate frames between two music performance videos. We found that the generated performance videos were of higher quality than the baseline method through quantitative experiments.

This paper proposes a light source selection for photon mapping combined with recent deep-learning-based importance sampling. Although applying such neural importance sampling (NIS) to photon mapping is not difficult, a straightforward approach can sample inappropriate photons for each light source because NIS relies on the approximation of a smooth continuous probability density function on the primary sample space, whereas the light source selection follows a discrete probability distribution. To alleviate this problem, we introduce a normalizing flow conditioned by a feature vector representing the index for each light source. When the neural network for NIS is trained to sample visible photons, we achieved lower variance with the same sample budgets, compared to a previous photon sampling using Markov chain Monte Carlo.

In scientific publications, information is conveyed in the form of figure and table as well as text. Among the many fields of research, computer vision focuses on visual information like image and video. Therefore, figure and table are useful to convey information in computer vision paper such as image used in the experiment or output of the proposed method. In the field of computer vision, conference papers are important, which is different from journal publications considered important in other fields. In this work, we study the use of figures and tables in computer vision papers in conference proceedings. We utilize object detection and image recognition techniques to extract and label figures in papers. We conducted the experiments from five aspects including (1) comparison with other field, (2) comparison among different conferences in computer vision, (3) comparison with workshop papers, (4) temporal change, and (5) comparison among different research topics in computer vision. Thorough the experiments, we observed that the use of figure and table has been changing between 2013 and 2020. We also revealed that the tendency in the use of figure is different among topics even in computer vision papers.

This paper presents a deep learning based method that generates body motion for string instrument performance from raw audio. In contrast to prior methods which aim to predict joint position from audio, we first estimate information that dictates the bowing dynamics, such as the bow direction and the played string. The final body motion is then determined from this information following a conversion rule. By adopting the bowing information as the target domain, not only is learning the mapping more feasible, but also the produced results have bowing dynamics that are consistent with the given audio. We confirmed that our results are superior to existing methods through extensive experiments.

Summaries of scientific publications enable readers to gain an overview of a large number of studies, but users’ preferences have not yet been explored. In this paper, we conduct two user studies (i.e., short-and long-term studies) where Japanese university students read summaries of English research articles that were either manually written or automatically generated using text summarization and/or machine translation. In the short-term experiment, subjects compared and evaluated the two types of summaries of the same article. We analyze the characteristics in the generated summaries that readers regard as important, such as content richness and simplicity. The experimental results show that subjects are mainly judged based on four criteria, including content richness, simplicity, fluency, and format. In the long-term experiment, subjects read 50 summaries and answered whether they would like to read the original papers after reading the summaries. We discuss the characteristics in the summaries that readers tend to use to determine whether to read the papers, such as topic, methods, and results. The comments from subjects indicate that specific components of scientific publications, including research topics and methods, are important to judge whether to read or not. Our study provides insights to enhance the effectiveness of automatic summarization of scientific publications.

This paper describes a representation learning method for disentangling an arbitrary musical instrument sound into latent pitch and timbre representations. Although such pitch-timbre disentanglement has been achieved with a variational autoencoder (VAE), especially for a predefined set of musical instruments, the latent pitch and timbre representations are outspread, making them hard to interpret. To mitigate this problem, we introduce a metric learning technique into a VAE with latent pitch and timbre spaces so that similar (different) pitches or timbres are mapped close to (far from) each other. Specifically, our VAE is trained with additional contrastive losses so that the latent distances between two arbitrary sounds of the same pitch or timbre are minimized, and those of different pitches or timbres are maximized. This training is performed under weak supervision that uses only whether the pitches and timbres of two sounds are the same or not, instead of their actual values. This improves the generalization capability for unseen musical instruments. Experimental results show that the proposed method can find better-structured disentangled representations with pitch and timbre clusters even for unseen musical instruments.

This paper introduces the 2nd place solution for the Riiid! Answer
Correctness Prediction in Kaggle, the world's largest data science competition
website. This competition was held from October 16, 2020, to January 7, 2021,
with 3395 teams and 4387 competitors. The main insights and contributions of
this paper are as follows. (i) We pointed out existing Transformer-based models
are suffering from a problem that the information which their query/key/value
can contain is limited. To solve this problem, we proposed a method that uses
LSTM to obtain query/key/value and verified its effectiveness. (ii) We pointed
out 'inter-container' leakage problem, which happens in datasets where
questions are sometimes served together. To solve this problem, we showed
special indexing/masking techniques that are useful when using RNN-variants and
Transformer. (iii) We found additional hand-crafted features are effective to
overcome the limits of Transformer, which can never consider the samples older
than the sequence length.

In this paper, we address the open question: "What do adversarially robust models look at?" Recently, it has been reported in many works that there exists the trade-off between standard accuracy and adversarial robustness. According to prior works, this trade-off is rooted in the fact that adversarially robust and standard accurate models might depend on very different sets of features. However, it has not been well studied what kind of difference actually exists. In this paper, we analyze this difference through various experiments visually and quantitatively. Experimental results show that adversarially robust models look at things at a larger scale than standard models and pay less attention to fine textures. Furthermore, although it has been claimed that adversarially robust features are not compatible with standard accuracy, there is even a positive effect by using them as pre-trained models particularly in low resolution datasets.

A major challenge in the field of education is providing review schedules
that present learned items at appropriate intervals to each student so that
memory is retained over time. In recent years, attempts have been made to
formulate item reviews as sequential decision-making problems to realize
adaptive instruction based on the knowledge state of students. It has been
reported previously that reinforcement learning can help realize mathematical
models of students learning strategies to maintain a high memory rate. However,
optimization using reinforcement learning requires a large number of
interactions, and thus it cannot be applied directly to actual students. In
this study, we propose a framework for optimizing teaching strategies by
constructing a virtual model of the student while minimizing the interaction
with the actual teaching target. In addition, we conducted an experiment
considering actual instructions using the mathematical model and confirmed that
the model performance is comparable to that of conventional teaching methods.
Our framework can directly substitute mathematical models used in experiments
with human students, and our results can serve as a buffer between theoretical
instructional optimization and practical applications in e-learning systems.

The automatic summarization of scientific papers, to assist researchers in conducting literature surveys, has garnered significant attention because of the rapid increase in the number of scientific articles published each year. However, whether and how these summaries actually help readers in comprehending scientific papers has not been examined yet. In this work, we study the effectiveness of automatically generated summaries of scientific papers for students who do not have sufficient knowledge in research. We asked six students, enrolled in bachelor’s and master’s programs in Japan, to prepare a presentation on a scientific paper by providing them either the article alone, or the article and its summary generated by an automatic summarization system, after 15 min of reading time. The comprehension of an article was judged by four evaluators based on the participant’s presentation. The experimental results show that the completeness of the comprehension of the four participants was higher overall when the summary of the paper was provided. In addition, four participants, including the two whose completeness score reduced when the summary was provided, mentioned that the summary is helpful to comprehend a research article within a limited time span.

The exponential growth of scientific literature yields the need to support users to both effectively and efficiently analyze and understand the some body of research work. This exploratory process can be facilitated by providing graphical abstracts-a visual summary of a scientific publication. Accordingly, previous work recently presented an initial study on automatic identification of a central figure in a scientific publication, to be used as the publication's visual summary. This study, however, have been limited only to a single (biomedical) domain. This is primarily because the current state-of-the-art relies on supervised machine learning, typically relying on the existence of large amounts of labeled data: the only existing annotated data set until now covered only the biomedical publications. In this work, we build a novel benchmark data set for visual summary identification from scientific publications, which consists of papers presented at conferences from several areas of computer science. We couple this contribution with a new self-supervised learning approach to learn a heuristic matching of in-text references to figures with figure captions. Our self-supervised pre-training, executed on a large unlabeled collection of publications, attenuates the need for large annotated data sets for visual summary identification and facilitates domain transfer for this task. We evaluate our self-supervised pretraining for visual summary identification on both the existing biomedical and our newly presented computer science data set. The experimental results suggest that the proposed method is able to outperform the previous state-of-the-art without any task-specific annotations.

We propose a learning-based method of estimating the compatibility between vocal and accompaniment audio tracks, i.e., how well they go with each other when played simultaneously. This task is challenging because it is difficult to formulate hand-crafted rules or construct a large labeled dataset to perform supervised learning. Our method uses self-supervised and joint-embedding techniques for estimating vocal-accompaniment compatibility. We train vocal and accompaniment encoders to learn a joint-embedding space of vocal and accompaniment tracks, where the embedded feature vectors of a compatible pair of vocal and accompaniment tracks lie close to each other and those of an incompatible pair lie far from each other. To address the lack of large labeled datasets consisting of compatible and incompatible pairs of vocal and accompaniment tracks, we propose generating such a dataset from songs using singing voice separation techniques, with which songs are separated into pairs of vocal and accompaniment tracks, and then original pairs are assumed to be compatible, and other random pairs are not. We achieved this training by constructing a large dataset containing 910,803 songs and evaluated the effectiveness of our method using ranking-based evaluation methods.

This paper proposes a statistical approach to 2D pose estimation from human images. The main problems with the standard supervised approach, which is based on a deep recognition (image-to-pose) model, are that it often yields anatomically implausible poses, and its performance is limited by the amount of paired data. To solve these problems, we propose a semi-supervised method that can make effective use of images with and without pose annotations. Specifically, we formulate a hierarchical generative model of poses and images by integrating a deep generative model of poses from pose features with that of images from poses and image features. We then introduce a deep recognition model that infers poses from images. Given images as observed data, these models can be trained jointly in a hierarchical variational autoencoding (image-to-pose-to-feature-to-pose-to-image) manner. The results of experiments show that the proposed reflective architecture makes estimated poses anatomically plausible, and the pose estimation performance is improved by integrating the recognition and generative models and also by feeding non-annotated images.

Providing visual summaries of scientific publications can increase information access for readers and thereby help deal with the exponential growth in the number of scientific publications. Nonetheless, efforts in providing visual publication summaries have been few and far apart, primarily focusing on the biomedical domain. This is primarily because of the limited availability of annotated gold standards, which hampers the application of robust and high-performing supervised learning techniques. To address these problems we create a new benchmark dataset for selecting figures to serve as visual summaries of publications based on their abstracts, covering several domains in computer science. Moreover, we develop a self-supervised learning approach, based on heuristic matching of inline references to figures with figure captions. Experiments in both biomedical and computer science domains show that our model is able to outperform the state of the art despite being self-supervised and therefore not relying on any annotated training data.

During the performance of sound source localization which uses both visual and aural information, it presently remains unclear how much either image or sound modalities contribute to the result, i.e. do we need both image and sound for sound source localization? To address this question, we develop an unsupervised learning system that solves sound source localization by decomposing this task into two steps: (i) “potential sound source localization”, a step that localizes possible sound sources using only visual information (ii) “object selection”, a step that identifies which objects are actually sounding using aural information. Our overall system achieves state-of-the-art performance in sound source localization, and more importantly, we find that despite the constraint on available information, the results of (i) achieve similar performance. From this observation and further experiments, we show that visual information is dominant in “sound” source localization when evaluated with the currently adopted benchmark dataset. Moreover, we show that the majority of sound-producing objects within the samples in this dataset can be inherently identified using only visual information, and thus that the dataset is inadequate to evaluate a system’s capability to leverage aural information. As an alternative, we present an evaluation protocol that enforces both visual and aural information to be leveraged, and verify this property through several experiments.

We present Song2Face, a deep neural network capable of producing singing facial animation from an input of singing voice and singer label. The network architecture is built upon our insight that, although facial expression when singing varies between different individuals, singing voices store valuable information such as pitch, breathe, and vibrato that expressions may be attributed to. Therefore, our network consists of an encoder that extracts relevant vocal features from audio, and a regression network conditioned on a singer label that predicts control parameters for facial animation. In contrast to prior audio-driven speech animation methods which initially map audio to text-level features, we show that vocal features can be directly learned from singing voice without any explicit constraints. Our network is capable of producing movements for all parts of the face and also rotational movement of the head itself. Furthermore, stylistic differences in expression between different singers are captured via the singer label, and thus the resulting animations singing style can be manipulated at test time.

We present a novel conceptaudio-visual object removalin 360-degree videos, in which a target object in a 360-degree video is removed in both the visual and auditory domains synchronously. Previous methods have solely focused on the visual aspect of object removal using video inpainting techniques, resulting in videos with unreasonable remaining sounds corresponding to the removed objects. We propose a solution which incorporates direction acquired during the video inpainting process into the audio removal process. More specifically, our method identifies the sound corresponding to the visually tracked target object and then synthesizes a three-dimensional sound field by subtracting the identified sound from the input 360-degree video. We conducted a user study showing that our multi-modal object removal supporting both visual and auditory domains could significantly improve the virtual reality experience, and our method could generate sufficiently synchronous, natural and satisfactory 360-degree videos.

Screen-space subsurface scattering is currently the most common approach to represent translucent materials in real-time rendering. However, most of the current approaches approximate the diffuse reflectance profile of translucent materials as a symmetric function, whereas the profile has an asymmetric shape in nature. To address this problem, we propose LinSSS, a numerical representation of heterogeneous subsurface scattering for real-time screen-space rendering. Although our representation is built upon a previous method, it makes two contributions. First, LinSSS formulates the diffuse reflectance profile as a linear combination of radially symmetric Gaussian functions. Nevertheless, it can also represent the spatial variation and the radial asymmetry of the profile. Second, since LinSSS is formulated using only the Gaussian functions, the convolution of the diffuse reflectance profile can be efficiently calculated in screen space. To further improve the efficiency, we deform the rendering equation obtained using LinSSS by factoring common convolution terms and approximate the convolution processes using a MIP map. Consequently, our method works as fast as the state-of-the-art method, while our method successfully represents the heterogeneity of scattering.

We present a guiding system to help blind people walk in public spaces while making their walking seamless with nearby pedestrians. Blind users carry a rolling suitcase-shaped system that has two RGBD Cameras, an inertial measurement unit (IMU) sensor, and light detection and ranging (LiDAR) sensor. The system senses the behavior of surrounding pedestrians, predicts risks of collisions, and alerts users to help them avoid collisions. It has two modes: The "on-path"mode that helps users avoid collisions without changing their path by adapting their walking speed; and the "off-path"mode that navigates an alternative path to go around pedestrians standing in the way Auditory and tactile modalities have been commonly used for non-visual navigation systems, so we implemented two interfaces to evaluate the effectiveness of each modality for collision avoidance. A user study with 14 blind participants in public spaces revealed that participants could successfully avoid collisions with both modalities. We detail the characteristics of each modality.

By overlaying virtual imagery onto the real world, mixed reality facilitates diverse applications and has drawn increasing attention. Enhancing physical in-hand objects with a virtual appearance is a key component for many applications that require users to interact with tools such as surgery simulations. However, due to complex hand articulations and severe hand-object occlusions, resolving occlusions in hand-object interactions is a challenging topic. Traditional tracking-based approaches are limited by strong ambiguities from occlusions and changing shapes, while reconstruction-based methods show a poor capability of handling dynamic scenes. In this article, we propose a novel real-time optimization system to resolve hand-object occlusions by spatially reconstructing the scene with estimated hand joints and masks. To acquire accurate results, we propose a joint learning process that shares information between two models and jointly estimates hand poses and semantic segmentation. To facilitate the joint learning system and improve its accuracy under occlusions, we propose an occlusion-aware RGB-D hand data set that mitigates the ambiguity through precise annotations and photorealistic appearance. Evaluations show more consistent overlays compared with literature, and a user study verifies a more realistic experience.

We present a novel method for simulating liquid with asynchronous time steps on Eulerian grids. Previous approaches focus on Smoothed Particle Hydrodynamics (SPH), Material Point Method (MPM) or tetrahedral Finite Element Method (FEM) but the method for simulating liquid purely on Eulerian grids have not yet been investigated. We address several challenges specifically arising from the Eulerian asynchronous time integrator such as regional pressure solve, asynchronous advection, interpolation, regional volume preservation, and dedicated segregation of the simulation domain according to the liquid velocity. We demonstrate our method on top of staggered grids combined with the level set method and the semi-Lagrangian scheme. We run several examples and show that our method considerably outperforms the global adaptive time step method with respect to the computational runtime on scenes where a large variance of velocity is present.

This paper proposes a statistical approach to 2D pose estimation from human
images. The main problems with the standard supervised approach, which is based
on a deep recognition (image-to-pose) model, are that it often yields
anatomically implausible poses, and its performance is limited by the amount of
paired data. To solve these problems, we propose a semi-supervised method that
can make effective use of images with and without pose annotations.
Specifically, we formulate a hierarchical generative model of poses and images
by integrating a deep generative model of poses from pose features with that of
images from poses and image features. We then introduce a deep recognition
model that infers poses from images. Given images as observed data, these
models can be trained jointly in a hierarchical variational autoencoding
(image-to-pose-to-feature-to-pose-to-image) manner. The results of experiments
show that the proposed reflective architecture makes estimated poses
anatomically plausible, and the performance of pose estimation improved by
integrating the recognition and generative models and also by feeding
non-annotated images.

Subsurface scattering involves the complicated behavior of light beneath the surfaces of translucent objects that includes scattering and absorption inside the object's volume. Physically accurate numerical representation of subsurface scattering requires a large number of parameters because of the complex nature of this phenomenon. The large amount of data restricts the use of the data on memory-limited devices such as video game consoles and mobile phones. To address this problem, this paper proposes an efficient data compression method for heterogeneous subsurface scattering. The key insight of this study is that heterogeneous materials often comprise a limited number of base materials, and the size of the subsurface scattering data can be significantly reduced by parameterizing only a few base materials. In the proposed compression method, we represent the scattering property of a base material using a function referred to as the base scattering profile. A small subset of the base materials is assigned to each surface position, and the local scattering property near the position is described using a linear combination of the base scattering profiles in the log scale. The proposed method reduces the data by a factor of approximately 30 compared to a state-of-the-art method, without significant loss of visual quality in the rendered graphics. In addition, the compressed data can also be used as bidirectional scattering surface reflectance distribution functions (BSSRDF) without incurring much computational overhead. These practical aspects of the proposed method also facilitate the use of higher-resolution BSSRDFs in devices with large memory capacity.

Modeling clothing to characters is one of the most time-consuming tasks for artists in 3DCG animation production. Transferring existing clothing models is a simple and powerful solution to reduce labor. In this paper, we propose a method to generate a clothing model for various characters from a single template model. Our framework consists of three steps: scale measurement, clothing transformation, and texture preservation. By introducing a novel measurement of the scale deviation between two characters with different shapes and poses, our framework achieves pose-independent transfer of clothing even for quadrupeds (e.g., from human to horse). In addition to a plausible clothing transformation method based on the scale measurement, our method minimizes texture distortion resulting from large deformation. We demonstrate that our system is robust for a wide range of body shapes and poses, which is challenging for current state-of-the-art methods.

HYPERMASK is a system which projects an animated face onto a physical mask, worn by an actor. As the mask moves within a prescribed area, its position and orientation are detected by a camera, and the projected image changes with respect to the viewpoint of the audience. The lips of the projected face are automatically synthesized in real time with the voice of the actor, who also controls the facial expressions. As a theatrical tool, HYPERMASK enables a new style of storytelling. As a prototype system, we propose to put a self-contained HYPERMASK system in a trolley (disguised as a linen cart), so that it projects onto the mask worn by the actor pushing the trolley.

With 360 imaging devices becoming widely accessible, omnidirectional content has gained popularity in multiple fields. The ability to estimate depth from a single omnidirectional image can benefit applications such as robotics navigation and virtual reality. However, existing depth estimation approaches produce sub-optimal results on real-world omnidirectional images with dynamic foreground objects. On the one hand, capture-based methods cannot obtain the foreground due to the limitations of the scanning and stitching schemes. On the other hand, it is challenging for synthesis-based methods to generate highly-realistic virtual foreground objects that are comparable to the real-world ones. In this paper, we propose to augment datasets with realistic foreground objects using an image-based approach, which produces a foreground-aware photorealistic dataset for machine learning algorithms. By exploiting a novel scale-invariant RGB-D correspondence in the spherical domain, we repurpose abundant non-omnidirectional datasets to include realistic foreground objects with correct distortions. We further propose a novel auxiliary deep neural network to estimate both the depth of the omnidirectional images and the mask of the foreground objects, where the two tasks facilitate each other. A new local depth loss considers small regions of interests and ensures that their depth estimations are not smoothed out during the global gradient’s optimization. We demonstrate the system using human as the foreground due to its complexity and contextual importance, while the framework can be generalized to any other foreground objects. Experimental results demonstrate more consistent global estimations and more accurate local estimations compared with state-of-the-arts.

This paper describes a method that generates in-between frames of two videos of a musical instrument being played. While image generation achieves a successful outcome in recent years, there is ample scope for improvement in video generation. The keys to improving the quality of video generation are the high resolution and temporal coherence of videos. We solved these requirements by using not only visual information but also aural information. The critical point of our method is using two-dimensional pose features to generate high-resolution in-between frames from the input audio. We constructed a deep neural network with a recurrent structure for inferring pose features from the input audio and an encoder-decoder network for padding and generating video frames using pose features. Our method, moreover, adopted a fusion approach of generating, padding, and retrieving video frames to improve the output video. Pose features played an essential role in both end-to-end training with a differentiable property and combining a generating, padding, and retrieving approach. We conducted a user study and confirmed that the proposed method is effective in generating interpolated videos.

Efficient car shape design is a challenging problem in both the automotive industry and the computer animation/games industry. In this paper, we present a system to reconstruct the 3D car shape from a single 2D sketch image. To learn the correlation between 2D sketches and 3D cars, we propose a Variational Autoencoder deep neural network that takes a 2D sketch and generates a set of multi-view depth and mask images, which form a more effective representation comparing to 3D meshes, and can be effectively fused to generate a 3D car shape. Since global models like deep learning have limited capacity to reconstruct fine-detail features, we propose a local lazy learning approach that constructs a small subspace based on a few relevant car samples in the database. Due to the small size of such a subspace, fine details can be represented effectively with a small number of parameters. With a low-cost optimization process, a high-quality car shape with detailed features is created. Experimental results show that the system performs consistently to create highly realistic cars of substantially different shape and topology.

We present a system to allow blind people to stand in line in public spaces by using an off-the-shelf smartphone only. The technologies to navigate blind pedestrians in public spaces are rapidly improving, but tasks which require to understand surrounding people's behavior are still difficult to assist. Standing in line at shops, stations, and other crowded places is one of such tasks. Therefore, we developed a system to detect and notify the distance to a person in front continuously by using a smartphone with a RGB camera and an infrared depth sensor. The system alerts three levels of distance via vibration patterns to allow users to start/stop moving forward to the right position at the right timing. To evaluate the effectiveness of the system, we performed a study with six blind people. We observed that the system enables blind participants to stand in line successfully, while also gaining more confidence.

Blind people face various local navigation challenges in their daily lives such as identifying empty seats in crowded stations, navigating toward a seat, and stopping and sitting at the correct spot. Although voice navigation is a commonly used solution, it requires users to carefully follow frequent navigational sounds over short distances. Therefore, we presented an assistive robot, BlindPilot, which guides blind users to landmark objects using an intuitive handle. BlindPilot employs an RGB-D camera to detect the positions of target objects and uses LiDAR to build a 2D map of the surrounding area. On the basis of the sensing results, BlindPilot then generates a path to the object and guides the user safely. To evaluate our system, we also implemented a sound-based navigation system as a baseline system, and asked six blind participants to approach an empty chair using the two systems. We observed that BlindPilot enabled users to approach a chair faster with a greater feeling of security and less effort compared to the baseline system.

Automatic dance synthesis has become more and more popular due to the increasing demand in computer games and animations. Existing research generates dance motions without much consideration for the context of the music. In reality, professional dancers make choreography according to the lyrics and music features. In this research, we focus on a particular genre of dance known as sign dance, which combines gesture-based sign language with full body dance motion. We propose a system to automatically generate sign dance from a piece of music and its corresponding sign gesture. The core of the system is a Sign Dance Model trained by multiple regression analysis to represent the correlations between sign dance and sign gesture/music, as well as a set of objective functions to evaluate the quality of the sign dance. Our system can be applied to music visualization, allowing people with hearing difficulties to understand and enjoy music.

Efficient car shape design is a challenging problem in both the automotive industry and the computer animation/games industry. In this paper, we present a system to reconstruct the 3D car shape from a single 2D sketch image. To learn the correlation between 2D sketches and 3D cars, we propose a Variational Autoencoder deep neural network that takes a 2D sketch and generates a set of multiview depth & mask images, which are more effective representation comparing to 3D mesh, and can be combined to form the 3D car shape. To ensure the volume and diversity of the training data, we propose a feature-preserving car mesh augmentation pipeline for data augmentation. Since deep learning has limited capacity to reconstruct fine-detail features, we propose a lazy learning approach that constructs a small subspace based on a few relevant car samples in the database. Due to the small size of such a subspace, fine details can be represented effectively with a small number of parameters. With a low-cost optimization process, a high-quality car with detailed features is created. Experimental results show that the system performs consistently to create highly realistic cars of substantially different shape and topology, with a very low computational cost.

Indirect illumination involving with visually rich participating media such as turbulent smoke and loud explosions contributes significantly to the appearances of other objects in a rendering scene. However, previous real-time techniques have focused only on the appearances of the media directly visible from the viewer. Specifically, appearances that can be indirectly seen over reflective surfaces have not attracted much attention. In this paper, we present a real-time rendering technique for such indirect views that involves the participating media. To achieve real-time performance for computing indirect views, we leverage layered polygonal area lights (LPALs) that can be obtained by slicing the media into multiple flat layers. Using this representation, radiance entering each surface point from each slice of the volume is analytically evaluated to achieve instant calculation. The analytic solution can be derived for standard bidirectional reflectance distribution functions (BRDFs) based on the microfacet theory. Accordingly, our method is sufficiently robust to work on surfaces with arbitrary shapes and roughness values. In addition, we propose a quadrature method for more accurate rendering of scenes with dense volumes, and a transformation of the domain of volumes to simplify the calculation and implementation of the proposed method. By taking advantage of these computation techniques, the proposed method achieves real-time rendering of indirect illumination for emissive volumes.

This paper presents an interactive face retrieval framework for clarifying an image representation envisioned by a user. Our system is designed for a situation in which the user wishes to find a person but has only visual memory of the person. We address a critical challenge of image retrieval across the user's inputs. Instead of target-specific information, the user can select several images that are similar to an impression of the target person the user wishes to search for. Based on the user's selection, our proposed system automatically updates a deep convolutional neural network. By interactively repeating these process, the system can reduce the gap between human-based similarities and computer-based similarities and estimate the target image representation. We ran user studies with 10 participants on a public database and confirmed that the proposed framework is effective for clarifying the image representation envisioned by the user easily and quickly.

This paper provides insight into the burden when people are walking up and down slopes in a virtual environment (VE) while actually walking on a flat floor in the real environment (RE). In RE, we feel a physical load during walking uphill or downhill. To reproduce such physical load in the VE, we provided visual stimuli to users by changing their step length. In order to investigate how the stimuli affect a sense of burden and body ownership, we performed a user study where participants walked on slopes in the VE. We found that changing the step length has a significant impact on a burden on the user and less correlation between body ownership and step length.

We developed an interactive music system called the "Melody Slot Machine," which provides an experience of manipulating a music performance. The melodies used in the system are divided into multiple segments, and each segment has multiple variations of melodies. By turning the dials manually, users can switch the variations of melodies freely. When you pull the slot lever, the melody of all segments rotates, and melody segments are randomly selected. Since the performer displayed in a hologram moves in accordance with the selected variation of melody, users can enjoy the feeling of manipulating the performance.

This paper presents a novel GPU-based algorithm for smoke animation. Our primary contribution is the use of Discrete Cosine Transform (DCT) compressed space for efficient simulation. We show that our method runs an order of magnitude faster than a CPU implementation while retaining visual details with a smaller memory usage. The key component of our method is an on-the-fly compression and expansion of velocity, pressure and density fields. Whenever these physical quantities are requested during a simulation, we perform data expansion and compression only where necessary in a loop. As a consequence, our simulation allows us to simulate a large domain without actually allocating full memory space for it. We show that albeit our method comes with some extra cost for DCT manipulations, such cost can be minimized with the aid of a devised shared memory usage.

We present an assistive suitcase system, BBeep, for supporting blind people when walking through crowded environments. BBeep uses pre-emptive sound notifications to help clear a path by alerting both the user and nearby pedestrians about the potential risk of collision. BBeep triggers notifications by tracking pedestrians, predicting their future position in real-time, and provides sound notifications only when it anticipates a future collision. We investigate how different types and timings of sound affect nearby pedestrian behavior. In our experiments, we found that sound emission timing has a significant impact on nearby pedestrian trajectories when compared to different sound types. Based on these findings, we performed a real-world user study at an international airport, where blind participants navigated with the suitcase in crowded areas. We observed that the proposed system significantly reduces the number of imminent collisions.

This paper proposes a lightweight bidirectional scattering distribution function (BSDF) model for layered materials with anisotropic reflection and refraction properties. In our method, each layer of the materials can be described by a microfacet BSDF using an anisotropic normal distribution function (NDF). Furthermore, the NDFs of layers can be defined on tangent vector fields, which differ from layer to layer. Our method is based on a previous study in which isotropic BSDFs are approximated by projecting them onto base planes. However, the adequateness of this previous work has not been well investigated for anisotropic BSDFs. In this paper, we demonstrate that the projection is also applicable to anisotropic BSDFs and that they can be approximated by elliptical distributions using covariance matrices.

This study describes a method that automatically generates a piano reduction score from the audio recordings of popular music while considering the musical structure. The generated score comprises both right- and left-hand piano parts, which reflect the melodies, chords, and rhythms extracted from the original audio signals. Generating such a reduction score from an audio recording is challenging because automatic music transcription is still considered to be inefficient when the input contains sounds from various instruments. Reflecting the long-term correlation structure behind similar repetitive bars is also challenging; further, previous methods have independently generated each bar. Our approach addresses the aforementioned issues by integrating musical analysis, especially structural analysis, with music generation. Our method extracts rhythmic features as well as melodies and chords from the input audio recording and reflects them in the score. To consider the long-term correlation between bars, we use similarity matrices, created for several acoustical features, as constraints. We further conduct a multivariate regression analysis to determine the acoustical features that represent the most valuable constraints for generating a musical structure. We have generated piano scores using our method and have observed that we can produce scores that differently balance between the ability to achieve rhythmic characteristics and the ability to obtain musical structures.

There is a demand for arranging music composed using multiple instruments for a solo piano because there are several pianists who wish to practice playing their favorite songs or music. Generally, the method used for piano arrangement entails reducing original notes to fit on a two-line staff. However, a fundamental solution that improves originality and playability in conjunction with score quality continues to elude approaches proposed by extant studies. Hence, the present study proposes a new approach to arranging a musical score for the piano by using four musical components, namely melody, chords, rhythm, and the number of notes that can be extracted from an original score. The proposed method involves inputting an original score and subsequently generating both right- and left-hand playing parts of piano scores. With respect to the right part, optional notes from a chord were added to the melody. With respect to the left part, appropriate accompaniments were selected from a database comprising pop musical piano scores. The selected accompaniments are considered to correspond to the impression of an original score. High-quality solo piano scores reflecting original characteristics were generated and considered as part of playability.

Face recognition research is one of the most active topics in computer vision (CV), and deep neural networks (DNN) are now filling the gap between human-level and computer-driven performance levels in face verification algorithms. However, although the performance gap appears to be narrowing in terms of accuracy-based expectations, a curious question has arisen; specifically, Face understanding of AI is really close to that of human? In the present study, in an effort to confirm the brain-driven concept, we conduct image-based detection, classification, and generation using an in-house created fake face database. This database has two configurations: (i) false positive face detections produced using both the Viola Jones (VJ) method and convolutional neural networks (CNN), and (ii) simulacra that have fundamental characteristics that resemble faces but are completely artificial. The results show a level of suggestive knowledge that indicates the continuing existence of a gap between the capabilities of recent vision-based face recognition algorithms and human-level performance. On a positive note, however, we have obtained knowledge that will advance the progress of face-understanding models.

Due to the recent boom in artificial intelligence (AI) research, including computer vision (CV), it has become impossible for researchers in these fields to keep up with the exponentially increasing number of manuscripts. In response to this situation, this paper proposes the paper summary generation (PSG) task using a simple but effective method to automatically generate an academic paper summary from raw PDF data. We realized PSG by combination of vision-based supervised components detector and language-based unsupervised important sentence extractor, which is applicable for a trained format of manuscripts. We show the quantitative evaluation of ability of simple vision-based components extraction, and the qualitative evaluation that our system can extract both visual item and sentence that are helpful for understanding. After processing via our PSG, the 979 manuscripts accepted by the Conference on Computer Vision and Pattern Recognition (CVPR) 2018 are available 1. It is believed that the proposed method will provide a better way for researchers to stay caught with important academic papers.

Due to the recent boom in artificial intelligence (AI) research, including
computer vision (CV), it has become impossible for researchers in these fields
to keep up with the exponentially increasing number of manuscripts. In response
to this situation, this paper proposes the paper summary generation (PSG) task
using a simple but effective method to automatically generate an academic paper
summary from raw PDF data. We realized PSG by combination of vision-based
supervised components detector and language-based unsupervised important
sentence extractor, which is applicable for a trained format of manuscripts. We
show the quantitative evaluation of ability of simple vision-based components
extraction, and the qualitative evaluation that our system can extract both
visual item and sentence that are helpful for understanding. After processing
via our PSG, the 979 manuscripts accepted by the Conference on Computer Vision
and Pattern Recognition (CVPR) 2018 are available. It is believed that the
proposed method will provide a better way for researchers to stay caught with
important academic papers.

Due to the need to interact with virtual objects, the hand-object interaction has become an important element in mixed reality (MR) applications. In this paper, we propose a novel approach to handle the occlusion of augmented 3D object manipulation with hands by exploiting the nature of hand poses combined with tracking-based and model-based methods, to achieve a complete mixed reality experience without necessities of heavy computations, complex manual segmentation processes or wearing special gloves. The experimental results show a frame rate faster than real-time and a great accuracy of rendered virtual appearances, and a user study verifies a more immersive experience compared to past approaches. We believe that the proposed method can improve a wide range of mixed reality applications that involve hand-object interactions.

This paper proposes efficient path sampling for re-rendering scenes after material editing. The proposed sampling is based on Metropolis light transport (MLT) and dedicates path samples more to pixels receiving greater changes by editing. To roughly estimate the amount of the changes in pixel values, we first render the difference between images before and after editing. In this step, we render the difference image directly rather than taking the difference of the images by separately rendering them. Then, we sample more paths for the pixels with larger difference values, and render the scene after editing following a recent approach using the control variates. As a result, we can obtain fine rendering results with a small number of path samples. We examined the proposed sampling with a range of scenes and demonstrated that it achieves lower estimation errors and variances over the state-of-the-art method.

Face recognition research is one of the most active topics in computer vision
(CV), and deep neural networks (DNN) are now filling the gap between
human-level and computer-driven performance levels in face verification
algorithms. However, although the performance gap appears to be narrowing in
terms of accuracy-based expectations, a curious question has arisen;
specifically, "Face understanding of AI is really close to that of human?" In
the present study, in an effort to confirm the brain-driven concept, we conduct
image-based detection, classification, and generation using an in-house created
fake face database. This database has two configurations: (i) false positive
face detections produced using both the Viola Jones (VJ) method and
convolutional neural networks (CNN), and (ii) simulacra that have fundamental
characteristics that resemble faces but are completely artificial. The results
show a level of suggestive knowledge that indicates the continuing existence of
a gap between the capabilities of recent vision-based face recognition
algorithms and human-level performance. On a positive note, however, we have
obtained knowledge that will advance the progress of face-understanding models.

This abstract introduces a generative neural network for face swapping and editing face images. We refer to this network as "region-separative generative adversarial network (RSGAN)". In existing deep generative models such as Variational autoencoder (VAE) and Generative adversarial network (GAN), training data must represent what the generative models synthesize. For example, image inpainting is achieved by training images with and without holes. However, it is difficult or even impossible to prepare a dataset which includes face images both before and after face swapping because faces of real people cannot be swapped without surgical operations. We tackle this problem by training the network so that it synthesizes synthesize a natural face image from an arbitrary pair of face and hair appearances. In addition to face swapping, the proposed network can be applied to other editing applications, such as visual attribute editing and random face parts synthesis.

This poster proposes a new deep generative model that we refer to as region separative generative adversarial network (RSGAN), which can achieve face swapping between arbitrary image pairs and can robustly perform the swapping compared to previous methods using 3DMM.

We present a deep learning-based technique to infer high-quality facial reflectance and geometry given a single unconstrained image of the subject, which may contain partial occlusions and arbitrary illumination conditions. The reconstructed high-resolution textures, which are generated in only a few seconds, include high-resolution skin surface reflectance maps, representing both the diffuse and specular albedo, and medium-and high-frequency displacement maps, thereby allowing us to render compelling digital avatars under novel lighting conditions. To extract this data, we train our deep neural networks with a high-quality skin reflectance and geometry database created with a state-of-the-art multi-view photometric stereo system using polarized gradient illumination. Given the raw facial texture map extracted from the input image, our neural networks synthesize complete reflectance and displacement maps, as well as complete missing regions caused by occlusions. The completed textures exhibit consistent quality throughout the face due to our network architecture, which propagates texture features from the visible region, resulting in high-fidelity details that are consistent with those seen in visible regions. We describe how this highly underconstrained problem is made tractable by dividing the full inference into smaller tasks, which are addressed by dedicated neural networks. We demonstrate the effectiveness of our network design with robust texture completion from images of faces that are largely occluded. With the inferred reflectance and geometry data, we demonstrate the rendering of high-fidelity 3D avatars from a variety of subjects captured under different lighting conditions. In addition, we perform evaluations demonstrating that our method can infer plausible facial reflectance and geometric details comparable to those obtained from high-end capture devices, and outperform alternative approaches that require only a single unconstrained input image.

This paper presents an interactive face retrieval framework for clarifying an image representation envisioned by a user. Our system is designed for a situation in which the user wishes to find a person but has only visual memory of the person. We address a critical challenge of image retrieval across the user's inputs. Instead of target-specific information, the user can select several images (or a single image) that are similar to an impression of the target person the user wishes to search for. Based on the user's selection, our proposed system automatically updates a deep convolutional neural network. By interactively repeating these process (human-in-the-loop optimization), the system can reduce the gap between human-based similarities and computer-based similarities and estimate the target image representation. We ran user studies with 10 subjects on a public database and confirmed that the proposed framework is effective for clarifying the image representation envisioned by the user easily and quickly.

We investigate the potential of music appreciation using spatial mapping techniques, which allow us to “place” audio sources in various locations within a physical space. We consider possible ways of this new appreciation style and list some design variables, such as how to define coordinate systems, how to show visually, and who to place the sound sources. We conducted an exploratory user study to examine how these design variables affect users’ music listening experiences. Based on our findings from the study, we discuss how we should develop systems that incorporate these design variables for music appreciation in the future.

Voxelization is a crucial process for many computer graphics applications such as collision detection, rendering of translucent objects, and global illumination. However, in some situations, although the mesh looks good, the voxelization result may be undesirable. In this paper, we describe a novel voxelization method that uses the graphics processing unit for surface voxelization. Our improvements on the voxelization algorithm can address a problem of state-of-the-art voxelization, which cannot deal with thin parts of the mesh object. We improve the quality of voxelization on both normal mediation and surface correction. Furthermore, we investigate our voxelization methods on indirect illumination, showing the improvement on the quality of real-time rendering.

In this paper, we present an integrated system for automatically generating
and editing face images through face swapping, attribute-based editing, and
random face parts synthesis. The proposed system is based on a deep neural
network that variationally learns the face and hair regions with large-scale
face image datasets. Different from conventional variational methods, the
proposed network represents the latent spaces individually for faces and hairs.
We refer to the proposed network as region-separative generative adversarial
network (RSGAN). The proposed network independently handles face and hair
appearances in the latent spaces, and then, face swapping is achieved by
replacing the latent-space representations of the faces, and reconstruct the
entire face image with them. This approach in the latent space robustly
performs face swapping even for images which the previous methods result in
failure due to inappropriate fitting or the 3D morphable models. In addition,
the proposed system can further edit face-swapped images with the same network
by manipulating visual attributes or by composing them with randomly generated
face or hair parts.

Copyright is held by the author/owner(s). This work presents the Dynamic Object Scanning (DO-Scanning), a novel interface that helps users browse long and untrimmed first-person videos quickly. The proposed interface offers users a small set of object cues generated automatically tailored to the context of a given video. Users choose which cue to highlight, and the interface in turn fast-forwards the video adaptively while keeping scenes with highlighted cues played at original speed. Our experimental results have revealed that the DO-Scanning arranged an efficient and compact set of cues, and this set of cues is useful for browsing a diverse set of first-person videos.

Copyright is held by the author/owner(s). This work presents the Dynamic Object Scanning (DO-Scanning), a novel interface that helps users browse long and untrimmed first-person videos quickly. The proposed interface offers users a small set of object cues generated automatically tailored to the context of a given video. Users choose which cue to highlight, and the interface in turn adaptively fast-forwards the video while keeping scenes with highlighted cues played at original speed. Our experimental results have revealed that the DO-Scanning has an efficient and compact set of cues arranged dynamically and this set of cues is useful for browsing a diverse set of first-person videos.

This work presents the Dynamic Object Scanning (DO- Scanning), a novel interface that helps users browse long and untrimmed first-person videos quickly. The proposed interface offers users a small set of object cues generated automatically tailored to the context of a given video. Users choose which cue to highlight, and the interface in turn adaptively fast-forwards the video while keeping scenes with highlighted cues played at original speed. Our experimental results have revealed that the DO-Scanning has an efficient and compact set of cues arranged dynamically and this set of cues is useful for browsing a diverse set of first-person videos.

In this paper, we present an integrated system for automatically generating and editing face images through face swapping, attribute-based editing, and random face parts synthesis. The proposed system is based on a deep neural network that variationally learns the face and hair regions with large-scale face image datasets. Different from conventional variational methods, the proposed network represents the latent spaces individually for faces and hairs. We refer to the proposed network as region-separative generative adversarial network (RSGAN). The proposed network independently handles face and hair appearances in the latent spaces, and then, face swapping is achieved by replacing the latent-space representations of the faces, and reconstruct the entire face image with them. This approach in the latent space robustly performs face swapping even for images which the previous methods result in failure due to inappropriate fitting or the 3D morphable models. In addition, the proposed system can further edit face-swapped images with the same network by manipulating visual attributes or by composing them with randomly generated face or hair parts.

Dance is an important component of live performance for expressing emotion and presenting visual context. Human dance performances typically require expert knowledge of dance choreography and professional rehearsal, which are too costly for casual entertainment venues and clubs. Recent advancements in character animation and motion synthesis have made it possible to synthesize virtual 3D dance characters in real-time. The major problem in existing systems is a lack of an intuitive interfaces to control the animation for real-time dance controls. We propose a new system called the DanceDJ to solve this problem. Our system consists of two parts. The first part is an underlying motion analysis system that evaluates motion features including dance features such as the postures and movement tempo, as well as audio features such as the music tempo and structure. As a pre-process, given a dancing motion database, our system evaluates the quality of possible timings to connect and switch different dancing motions. During run-time, we propose a control interface that provides visual guidance. We observe that disk jockeys (DJs) effectively control the mixing of music using the DJ controller, and therefore propose a DJ controller for controlling dancing characters. This allows DJs to transfer their skills from music control to dance control using a similar hardware setup. We map different motion control functions onto the DJ controller, and visualize the timing of natural connection points, such that the DJ can effectively govern the synthesized dance motion. We conducted two user experiments to evaluate the user experience and the quality of the dance character. Quantitative analysis shows that our system performs well in both motion control and simulation quality.

© Springer International Publishing AG, part of Springer Nature 2018. Limited animation is one of the traditional techniques for producing cartoon animations. Owing to its expressive style, it has been enjoyed around the world. However, producing high quality animations using this limited style is time-consuming and costly for animators. Furthermore, proper synchronization between the voice-actor’s voice and the character’s mouth and lip motion requires well-experienced animators. This is essential because viewers are very sensitive to audio-lip discrepancies. In this paper, we propose a method that automatically creates high-quality limited-style lip-synched animations using audio tracks. Our system can be applied for creating not only the original animations but also dubbed ones independently of languages. Because our approach follows the standard workflow employed in cartoon animation production, our system can successfully assist animators. In addition, users can implement our system as a plug-in of a standard tool for creating animations (Adobe After Effects) and can easily arrange character lip motion to suit their own style. We visually evaluate our results both absolutely and relatively by comparing them with those of previous works. From the user evaluations, we confirm that our algorithms is able to successfully generate more natural audio-mouth synchronizations in limited-style lip-synched animations than previous algorithms.

Due to the need to interact with virtual objects, the hand-object interaction has become an important element in mixed reality (MR) applications. In this paper, we propose a novel approach to handle the occlusion of augmented 3D object manipulation with hands by exploiting the nature of hand poses combined with tracking-based and model-based methods, to achieve a complete mixed reality experience without necessities of heavy computations, complex manual segmentation processes or wearing special gloves. The experimental results show a frame rate faster than real-time and a great accuracy of rendered virtual appearances, and a user study verifies a more imnersive experience compared to past approaclies. We believe that the proposed method can improve a wide range of mixed reality applications that involve hand-object interactions.

Copyright © 2018 by SCITEPRESS - Science and Technology Publications, Lda. All rights reserved. Aging of a human face is accompanied by visible changes such as sagging, spots, somberness, and wrinkles. Age progression techniques that estimate an aged facial image are required for long-term criminal or missing person investigations, and also in 3DCG facial animations. This paper focuses on aged facial texture and introduces a novel age progression method based on medical knowledge, which represents an aged wrinkles shapes and positions individuality. The effectiveness of the idea including expression wrinkles in aging facial image synthesis is confirmed through subjective evaluation.

This paper presents an interactive face retrieval framework for clarifying an image representation envisioned by a user. Our system is designed for a situation in which the user wishes to find a person but has only visual memory of the person. We address a critical challenge of image retrieval across the user's inputs. Instead of target-specific information, the user can select several images (or a single image) that are similar to an impression of the target person the user wishes to search for. Based on the user's selection, our proposed system automatically updates a deep convolutional neural network. By interactively repeating these process (human-in-the-loop optimization), the system can reduce the gap between human-based similarities and computer-based similarities and estimate the target image representation. We ran user studies with 10 subjects on a public database and con firmed that the proposed framework is effective for clarifying the image representation envisioned by the user easily and quickly.

We investigate the potential of music appreciation using spatial mapping techniques, which allow us to "place" audio sources in various locations within a physical space. We consider possible ways of this new appreciation style and list some design variables, such as how to define coordinate systems, how to show visually, and how to place the sound sources. We conducted an exploratory user study to examine how these design variables affect users' music listening experiences. Based on our findings from the study, we discuss how we should develop systems that incorporate these design variables for music appreciation in the future.

© 2018 Information Processing Society of Japan. This paper presents a topic modeling method to retrieve similar music fragments and its application, Music- Mixer, which is a computer-aided DJ system that supports DJ performance by automatically mixing songs in a seamless manner. MusicMixer mixes songs based on audio similarity calculated via beat analysis and latent topic analysis of the chromatic signal in the audio. The topic represents latent semantics on how chromatic sounds are generated. Given a list of songs, a DJ selects a song with beats and sounds similar to a specific point of the currently playing song to seamlessly transition between songs. By calculating similarities between all existing song sections that can be naturally mixed, MusicMixer retrieves the best mixing point from a myriad of possibilities and enables seamless song transitions. Although it is comparatively easy to calculate beat similarity from audio signals, considering the semantics of songs from the viewpoint of a human DJ has proven difficult. Therefore, we propose a method to represent audio signals to construct topic models that acquire latent semantics of audio. The results of a subjective experiment demonstrate the effectiveness of the proposed latent semantic analysis method. MusicMixer achieves automatic song mixing using the audio signal processing approach; thus, users can perform DJ mixing simply by selecting a song from a list of songs suggested by the system.

Limited animation is one of the traditional techniques for producing cartoon animations. Owing to its expressive style, it has been enjoyed around the world. However, producing high quality animations using this limited style is time-consuming and costly for animators. Furthermore, proper synchronization between the voice-actor's voice and the character's mouth and lip motion requires well-experienced animators. This is essential because viewers are very sensitive to audio-lip discrepancies. In this paper, we propose a method that automatically creates high-quality limited-style lip-synched animations using audio tracks. Our system can be applied for creating not only the original animations but also dubbed ones independently of languages. Because our approach follows the standard workflow employed in cartoon animation production, our system can successfully assist animators. In addition, users can implement our system as a plug-in of a standard tool for creating animations (Adobe After Effects) and can easily arrange character lip motion to suit their own style. We visually evaluate our results both absolutely and relatively by comparing them with those of previous works. From the user evaluations, we confirm that our algorithms is able to successfully generate more natural audio-mouth synchronizations in limited-style lipsynched animations than previous algorithms.

In recent years, a large number of makeup images have been shared on social media. Most of these images lack information about the cosmetics used, such as color, glitter or etc., while they are difficult to infer due to the diversity of skin color or lighting conditions. In this paper, our goal is to estimate cosmetic features only from a single makeup image. Previous work has measured the material parameters of cosmetic products from pairs of images showing the face with and without makeup, but such comparison images are not always available. Furthermore, this method cannot represent local effects such as pearl or glitter since they adapted physically-based reflectance models. We propose a novel image-based method to extract cosmetic features considering both color and local effects by decomposing the target image into makeup and skin color using Difference of Gaussian (DoG). Our method can be applied for single, standalone makeup images, and considers both local effects and color. In addition, our method is robust to the skin color difference due to the decomposition separating makeup from skin. The experimental results demonstrate that our method is more robust to skin color difference and captures characteristics of each cosmetic product.

This paper presents a topic modeling method to retrieve similar music fragments and its application, Music- Mixer, which is a computer-aided DJ system that supports DJ performance by automatically mixing songs in a seamless manner. MusicMixer mixes songs based on audio similarity calculated via beat analysis and latent topic analysis of the chromatic signal in the audio. The topic represents latent semantics on how chromatic sounds are generated. Given a list of songs, a DJ selects a song with beats and sounds similar to a specific point of the currently playing song to seamlessly transition between songs. By calculating similarities between all existing song sections that can be naturally mixed, MusicMixer retrieves the best mixing point from a myriad of possibilities and enables seamless song transitions. Although it is comparatively easy to calculate beat similarity from audio signals, considering the semantics of songs from the viewpoint of a human DJ has proven difficult. Therefore, we propose a method to represent audio signals to construct topic models that acquire latent semantics of audio. The results of a subjective experiment demonstrate the effectiveness of the proposed latent semantic analysis method. MusicMixer achieves automatic song mixing using the audio signal processing approach
thus, users can perform DJ mixing simply by selecting a song from a list of songs suggested by the system.

Dance is an important component of live performance for expressing emotion and presenting visual context. Human dance performances typically require expert knowledge of dance choreography and professional rehearsal, which are too costly for casual entertainment venues and clubs. Recent advancements in character animation and motion synthesis have made it possible to synthesize virtual 3D dance characters in real-time. The major problem in existing systems is a lack of an intuitive interfaces to control the animation for real-time dance controls. We propose a new system called the DanceDJ to solve this problem. Our system consists of two parts. The first part is an underlying motion analysis system that evaluates motion features including dance features such as the postures and movement tempo, as well as audio features such as the music tempo and structure. As a pre-process, given a dancing motion database, our system evaluates the quality of possible timings to connect and switch different dancing motions. During run-time, we propose a control interface that provides visual guidance. We observe that disk jockeys (DJs) effectively control the mixing of music using the DJ controller, and therefore propose a DJ controller for controlling dancing characters. This allows DJs to transfer their skills from music control to dance control using a similar hardware setup. We map different motion control functions onto the DJ controller, and visualize the timing of natural connection points, such that the DJ can effectively govern the synthesized dance motion. We conducted two user experiments to evaluate the user experience and the quality of the dance character. Quantitative analysis shows that our system performs well in both motion control and simulation quality.

' 2017 ACM. In this paper, we present an interactive framework to model garments for animals from a template garment model based on correspondences between the source and the target bodies. We address two critical challenges of garment transfer across significantly different body shapes and postures (e.g., for quadruped and human); (1) ambiguity in the correspondences and (2) distortion due to large variation in scale of each body part. Our efficient cross-parameterization algorithm and intuitive user interface allow us to interactively compute correspondences and transfer the overall shape of garments. We also introduce a novel algorithm for local coordinate optimization that minimizes the distortion of transferred garments, which leads a resulting model to a quasi-developable surface and hence ready for fabrication. Finally, we demonstrate the robustness and effectiveness of our approach on a various garments and body shapes, showing that visually pleasant garment models for animals can be generated and fabricated by our system with minimal effort.

© 2017 ACM. Accurately tracking a Head Mounted Display (HMD) with a 6 degree of freedom is essential to achieve a comfortable and a nausea free experience in Virtual Reality. Existing commercial HMD systems using synchronized Infrared (IR) camera and blinking IR-LEDs can achieve highly accurate tracking. However, most of the o-the-shelf cameras do not support frame synchronization. In this paper, we propose a novel method for real time HMD pose estimation without using any camera synchronization or LED blinking. We extended over the state of the art pose estimation algorithm by introducing geometrically constrained optimization. In addition, we propose a novel system to increase robustness to the blurred IR-LEDs paerns appearing at high-velocity movements. The quantitative evaluations showed signicant improvements in pose stability and accuracy over wide rotational movements as well as a decrease in runtime.

Synthesizing a friction sound of deformable objects by a computer is challenging. We propose a novel physics-based approach to synthesize friction sounds based on dynamics simulation. In this work, we calculate the elastic deformation of an object surface when the object comes in contact with other objects. The principle of our method is to divide an object surface into microrectangles. The deformation of each microrectangle is set using two assumptions: the size of a microrectangle (1) changes by contacting other object and (2) obeys a normal distribution. We consider the sound pressure distribution and its space spread, consisting of vibrations of all microrectangles, to synthesize a friction sound at an observation point. We express the global motions of an object by position based dynamics where we add an adhesion constraint. Our proposed method enables the generation of friction sounds of objects in different materials by regulating the initial value of microrectangular parameters.

We propose a novel font called Beautifying Font to assist learning techniques in writing Chinese calligraphy. Chinese calligraphy has various expressions but they are hard to acquire for beginners. Beautifying Font visualizes the speed and pressure of brush-strokes so that users can intuitively understand how to write.

We propose a rally-rank evaluation based on the court transition information for volleyball video summarization considering the contents of the game. Our method uses the court transition information instead of non-robust visual features such as the position of a ball and players. Experimental results demonstrate the effectiveness that our method reflects viewers' preferences over previous methods.

© 2017 Copyright held by the owner/author(s). We propose a novel font called Beautifying Font to assist learning techniques in writing Chinese calligraphy. Chinese calligraphy has various expressions but they are hard to acquire for beginners. Beautifying Font visualizes the speed and pressure of brush-strokes so that users can intuitively understand how to write.

© 2017 Copyright held by the owner/author(s). Resizing of 3D model is necessary for computer graphics animation and application such as games and movies. In general, when users deform a target model, they built on a bounding box or a closed polygon mesh (cage) to enclose a target model. Then, the resizing is done by deforming the cage with target model. However, these approaches are not good for detailed adjustment of 3D shape because they do not preserve local information. In contrast, based on a local information (e.g., edge set and weight map), Sorkine et al. [Sorkine and Alexa 2007; Sorkine et al. 2004] can generate smooth and conformal deformation results with only a few control points. While these approaches are useful for some situations, the results depend on resolution and topology of the target model. In addition, these approaches do not consider texture (UV) information.

© 2017 Copyright held by the owner/author(s). We propose a rally-rank evaluation based on the court transition information for volleyball video summarization considering the contents of the game. Our method uses the court transition information instead of non-robust visual features such as the position of a ball and players. Experimental results demonstrate the effectiveness that our method reflects viewers' preferences over previous methods.

Resizing of 3D model is necessary for computer graphics animation and application such as games and movies. In general, when users deform a target model, they built on a bounding box or a closed polygon mesh (cage) to enclose a target model. Then, the resizing is done by deforming the cage with target model. However, these approaches are not good for detailed adjustment of 3D shape because they do not preserve local information. In contrast, based on a local information (e.g., edge set and weight map), Sorkine et al. [Sorkine and Alexa 2007
Sorkine et al. 2004] can generate smooth and conformal deformation results with only a few control points. While these approaches are useful for some situations, the results depend on resolution and topology of the target model. In addition, these approaches do not consider texture (UV) information.

© 2017 Copyright held by the owner/author(s). Marker-based retexturing is to superimpose the texture on a target object by detecting and identifying markers from within the captured image. We propose a new marker that can be identified under a large deformation that involves self-occlusion, which was not taken into consideration in the following markers. Bradley et al. [Bradley et al. 2009] designed the independent markers, but it is difficult to recognize them under complicated occlusion. Scholz et al.[Scholz and Magnor 2006] created a circular marker with a single color selected from multiple colors. They created ID corresponding to the alignment of colors by one marker and the markers around it and identified by placing the marker so that the ID would be unique. However, when some markers are covered by self-occlusion, the positional relationship of the markers appears to be different from the original, so markers near the self-occlusion are failed to identify. Narita et al. [Narita et al. 2017] considered self-occlusion by improving the identification algorithm. They succeeded in improving the accuracy of identification by creating triangle meshes whose vertices are the center of gravity of markers and assuming that they are close to a right isosceles triangle. However, since outliers are removed using angles, identification of markers may fail in the case of the object that is likely to be deformed in the shear direction like a cloth. Therefore, we considered self-occlusion by designing hierarchical markers so that they can be refferred to in a global scope. We designed a color based marker for easy recognition even at low resolution.

This paper proposes an age progression method for facial videos. Age is one of the main factors that changes the appearance of the face, due to the associated sagging, spots, and wrinkles. These aging features change in appearance depending on facial expressions. As an example, we see wrinkles appear in the face of the young when smiling, but the shape of wrinkles changes in older faces. Previous work has not considered the temporal changes of the face, using only static images as databases. To solve this problem, we extend the texture synthesis approach to use facial videos as source videos. First, we spatio-temporally align the videos of database to match the sequence of a target video. Then, we synthesize an aging face and apply the temporal changes of the target age to the wrinkles appearing in the facial expression image in the target video. As a result, our method successfully generates expression changes specific to the target age.

© 2017 ACM. This paper proposes an age progression method for facial videos. Age is one of the main factors that changes the appearance of the face, due to the associated sagging, spots, and wrinkles. These aging features change in appearance depending on facial expressions. As an example, we see wrinkles appear in the face of the young when smiling, but the shape of wrinkles changes in older faces. Previous work has not considered the temporal changes of the face, using only static images as databases. To solve this problem, we extend the texture synthesis approach to use facial videos as source videos. First, we spatio-temporally align the videos of database to match the sequence of a target video. Then, we synthesize an aging face and apply the temporal changes of the target age to the wrinkles appearing in the facial expression image in the target video. As a result, our method successfully generates expression changes specific to the target age.

本研究では，老化に伴う表情変化の違いを考慮した経年変化顔の動画像合成を行う．動画中の人物の経年変化は，映画のような映像作品における年齢変化表現において重要である．老化に伴って発生する顔特徴の一つである皺に注目すると，表情変化に伴う皺の変化は年齢に大きく依存する．静止画を対象とした経年変化顔合成手法は数多く存在するが，いずれも皺の動的な変化に焦点を置いていないため，年齢による表情変化の違いを表現することができない．そこで本研究では，データベースとして目標年代の人物の表情変化動画を用いることにより，表情変化に伴う皺の動的な変化を表現する．具体的には，入力動画の各表情に対して，データベースの類似表情を用いて，老化時の顔の構築を行う．また，対象人物の皺の深さを，目標年代の変化と一致するように操作する．以上により，表情変化による皺の動的変化を考慮した経年変化顔の動画合成を実現した．

本稿では，衣服のモデリングにおける労力削減を目的とし，テンプレートとなる１つの衣服モデルから，任意のキャラクタの体型に適合した衣服モデルと型紙を生成する手法を提案する． 生成される衣服の概形を確認しながらソースの身体とターゲット身体の対応関係をインタラクティブに修正できるインターフェースの導入により，身体のモデルの頂点数や頂点間の接続情報に依存しない衣服転写を実現する． また，ソースの衣服の形状を反映する最適化処理を行った後に可展面近似を行うことで，ソースの衣服のデザインから形状が大きく離れることを防ぎ，もっともらしい衣服モデルを生成することを実現する． 提案手法は生成した衣服モデルの型紙を出力することが可能なため，例えば人間とペットのペアルックの衣服制作など，現実世界におけるものづくり支援への応用が期待される．

Dance animations of a three-dimensional (3D) character rendered with computer graphics (CG) have been created by using motion capture systems or 3DCG animation editing tools. Since these methods require skills and a large amount of work from the creator, automated choreography has been developed to make synthesizing dance motions much easier. However, due to the limited variations of the results, users could not reflect their preferences to the output. Therefore, supporting the user’s preference in dance animation is still important.<br />
<br />
We propose a novel dance creation system which supports the user to create choreography with their preferences. A user first selects a preferred motion from the database, and then assigns it to an arbitrary section of the music. With a few mouse clicks, our system helps a user search for alternative dance motion that reflect his or<br />
her preference by using relevance feedback. The system automatically synthesizes a continuous choreography with the constraint condition that the motions specified by the user are fixed. We conducted user studies and found that users could create new dance motions easily with their preferences by using the system.

Creating secondary motion of character animation including jiggling of fat is demanded in the computer animation. In general, secondary motion from the primary motion of the character are expressed based on shape matching approaches. However, the previous methods do not account for the directional stretch characteristics and local stiffness at the same time, that is problematic to represent the effect of anatomical structure such as muscle fiber. Our framework allows user to edit the anatomical structure of the character model corresponding to creature’s body containing muscle and fat from the tetrahedral model and bone motion. Our method then simulates the elastic deformation considering anatomical structures defined directional stretch characteristics and stiffness on each layer. In addition, our method can add the constraint for local deformation (e.g. biceps) considering defined model’s characteristics.

Screen space ambient occlusion is a very efficient means to capture the shadows caused by adjacent objects. However it is incapable of expressing transparency of objects. We introduce an approach which behaves like the combination of ambient occlusion and translucency. This method is an extension of the traditional screen space ambient occlusion algorithm with extra density field input. It can be applied on rendering mesh objects, and moreover it is very suitable for rendering complex hair models. We use the new algorithm to approximate light attenuation though semi-transparent hairs at real-time. Our method is implemented on common GPU, and independent from pre-computation. When it is used in environment lighting, the hair shading is visually similar to however one order of magnitude faster than existing algorithm.

© Springer International Publishing AG 2017. Screen space ambient occlusion is a very efficient means to capture the shadows caused by adjacent objects. However it is incapable of expressing transparency of objects. We introduce an approach which behaves like the combination of ambient occlusion and translucency. This method is an extension of the traditional screen space ambient occlusion algorithm with extra density field input. It can be applied on rendering mesh objects, and moreover it is very suitable for rendering complex hair models. We use the new algorithm to approximate light attenuation though semi-transparent hairs at real-time. Our method is implemented on common GPU, and independent from pre-computation. When it is used in environment lighting, the hair shading is visually similar to however one order of magnitude faster than existing algorithm.

This paper presents a subtitle placement method that reduces unnecessary eye movements. Although methods that vary the position of subtitles have been discussed in a previous study, subtitles may overlap the region of interest (ROI). Therefore, we propose a dynamic subtitling method that utilizes eye-tracking data to avoid the subtitles from overlapping with important regions. The proposed method calculates the ROI based on the eye-tracking data of multiple viewers. By positioning subtitles immediately under the ROI, the subtitles do not overlap the ROI. Furthermore, we detect speakers in a scene based on audio and visual information to help viewers recognize the speaker by positioning subtitles near the speaker. Experimental results show that the proposed method enables viewers to watch the ROI and the subtitle in longer duration than traditional subtitles, and is effective in terms of enhancing the comfort and utility of the viewing experience.

In this paper, we propose a video summarization system for volleyball videos. Our system automatically detects rally scenes as self-consumable video segments and evaluates rally-rank for each rally scene to decide priority. In the priority decision, features representing the contents of the game are necessary; however such features have not been considered in most previous methods. Although several visual features such as the position of a ball and players should be used, acquisition of such features is still non-robust and unreliable in low resolution or low frame rate volleyball videos. Instead, we utilize the court transition information caused by camera operation. Experimental results demonstrate the robustness of our rally scene detection and the effectiveness of our rally-rank to reflect viewers' preferences over previous methods.

Accurately tracking a Head Mounted Display (HMD) with a 6 degree of freedom is essential to achieve a comfortable and a nausea free experience in Virtual Reality. Existing commercial HMD systems using synchronized Infrared (IR) camera and blinking IR-LEDs can achieve highly accurate tracking. However, most of the off-the-shelf cameras do not support frame synchronization. In this paper, we propose a novel method for real time HMD pose estimation without using any camera synchronization or LED blinking. We extended over the state of the art pose estimation algorithm by introducing geometrically constrained optimization. In addition, we propose a novel system to increase robustness to the blurred IR-LEDs patterns appearing at high-velocity movements. The quantitative evaluations showed significant improvements in pose stability and accuracy over wide rotational movements as well as a decrease in runtime.

© 2017 Institute of Image Electronics Engineers of Japan. All rights reserved. In this paper, we introduce a technique of facial fattening (or slimming) simulation from a frontal facial image. In previous work, data-driven approaches enable a rapid design of fattening and slimming images that reflect realistic facial shapes; however, they focus only on facial surface such as facial landmarks, so it is difficult to consider any anatomical model such as skull structure. Then, we construct a novel facial database, which contains sets of frontal facial image and MRI image. Based on our database, we estimate 2D skull structure from the input image, and define a fattening (or slimming) rule and a skull-based constraint. As a result, we can generate a natural deformed image for fattening and slimming simulation.

© 2017 Institute of Image Electronics Engineers of Japan. All rights reserved. To synthesize photo-realistic images, simulating the subsurface scattering is highly important. However, physically accurate simulation of the subsurface scattering requires computationally complex integration over various optical paths going through translucent materials. On the other hand, the powers of light rays are immediately decreased when they pass through limitedly translucent materials. We leverage this observation to compute the light contribution approximately by only considering a single ray that has the most dominant contribution. The proposed method is based on an empirical assumption where the light ray having passed such a dominant path has the largest influence for the final rendering result. By only considering the single dominant path to evaluate the radiance, visually plausible rendering results can be obtained in real-time, while the rendering process itself is not based on the strict physics, nevertheless. This lightweight rendering process is particularly important for inhomogeneous translucent objects that include different translucent materials inside, which have been one of the targets that is difficult to render in real-time due to its high computational complexity.

We have a desire to edit images for various purposes such as art, entertainment, and film production so texture synthesis methods have been proposed. Especially, PatchMatch algorithm [Barnes et al.2009] enabled us to easily use many image editing tools. However, these tools are applied to one image. If we can automatically synthesize from various examples, we can create new and higher quality images. Visio-lization [Mohammed et al. 2009] generated average face by synthesis of face image database. However, the synthesis was applied block-wise so there were artifacts on the result and free form features of source images such as wrinkles could not be preserved. We proposed a new synthesis method for multiple images. We applied sparse and dense nearest neighbor search so that we can preserve both input and source database image features. Our method allows us to create a novel image from a number of examples.

© 2016 ACM. We have a desire to edit images for various purposes such as art, entertainment, and film production so texture synthesis methods have been proposed. Especially, PatchMatch algorithm [Barnes et al. 2009] enabled us to easily use many image editing tools. However, these tools are applied to one image. If we can automatically synthesize from various examples, we can create new and higher quality images. Visio-lization [Mohammed et al. 2009] generated average face by synthesis of face image database. However, the synthesis was applied block-wise so there were artifacts on the result and free form features of source images such as wrinkles could not be preserved. We proposed a new synthesis method for multiple images. We applied sparse and dense nearest neighbor search so that we can preserve both input and source database image features. Our method allows us to create a novel image from a number of examples.

3D facial shape reconstruction in the wild environments is an important research task in the field of CG and CV. This is because it can be applied to a lot of products, such as 3DCG video games and face recognition. One of the most popular 3D facial shape reconstruction techniques is 3D Model-based approach. This approach approximates a facial shape by using 3D face model, which is calculated by principal component analysis. [Blanz and Vetter 1999] performed a 3D facial reconstruction by fitting points from facial feature points of an input of single facial image to vertex of template 3D facial model named 3D Morphable Model. This method can reconstruct a facial shape from a variety of images which include different lighting and face orientation, as long as facial feature points can be detected. However, representation quality of the result depends on the number of 3D model resolution.

In recent years, thanks to the development of 3DCG animation editing tools (e.g. MikuMikuDance), a lot of 3D character dance animation movies are created by amateur users. However it is very difficult to create choreography from scratch without any technical knowledge. Shiratori et al. [2006] produced the dance automatic generation system considering rhythm and intensity of dance motions. However each segment is selected randomly from database, so the generated dance motion has no linguistic or emotional meanings. Takano et al. [2010] produced a human motion generation system considering motion labels. However they use simple motion labels like "running" or "jump", so they cannot generate motions that express emotions. In reality, professional dancers make choreography based on music features or lyrics in music, and express emotion or how they feel in music. In our work, we aim at generating more emotional dance motion easily. Therefore, we use linguistic information in lyrics, and generate dance motion.

Numerous video have been translated using "dubbing," spurred by the recent growth of video market. However, it is very difficult to achieve the visual-audio synchronization. That is to say in general a new audio does not synchronize with actor's mouth motion. This discrepancy can disturb comprehension of video contents. There-fore many methods have been researched so far to solve this problem.

© 2016 Copyright held by the owner/author(s). This paper presents a video-based method for detecting drowsiness. Generally, human beings can perceive their fatigue and drowsiness through looking at faces. The ability to perceive the fatigue and the drowsiness has been studied in many ways. The drowsiness detection method based on facial videos has been proposed [Nakamura et al. 2014]. In their method, a set of the facial features calculated with the Computer Vision techniques and the k-nearest neighbor algorithm are applied to classify drowsiness degree. However, the facial features that are ineffective against reproducing the perception of human beings with the machine learning method are not removed. This factor can decrease the detection accuracy.

An appearance of a human face changes with aging: sagging, spots, somberness, and wrinkles would be observed. Considering these changes, aging simulation techniques that estimate an aged facial image is required for a long-term criminal or missing person investigation. One of the latest works proposed a photorealistic aging simulation method using a patch-based facial image reconstruction. However, including this method, the latest works had a problem that they cannot represent an individuality of wrinkles which is one of the most important features that represent the human individuality. The individuality of wrinkles is defined by the shape and the position. In this paper, we introduce a novel aging simulation method with patch-based facial image reconstruction, which could overcome problem mentioned above. To preserve a wrinkles individuality, wrinkles which is in an expressive facial image is synthesized to an input image based on a medical knowledge. Our method represents the individuality of wrinkles, and subjective evaluation results describe that our method generates more accurate results when compared our work and the previous work with the ground truth.

Copyright is held by the owner/author(s). We present a subtitle-placement method that reduces viewer's eye movement without interfering with the target region of interest (ROI) in a video scene. Subtitles help viewers understand foreign-language videos. However, subtitles tend to attract viewers' line of sight, which cause viewers to lose focus on the video content. To address this problem, previous studies have attempted to improve viewer experiences by dynamically shifting subtitle positions. Nevertheless, in their user studies, some participants felt that the visual appearance of such subtitles was unnatural and caused them fatigue. We propose a method that places subtitles below the ROI, which is calculated by eye-tracking data from multiple viewers. Two experiments were conducted to evaluate viewer impression and compare line of sight for videos with subtitles placed by the proposed and previous methods.

Acquiring scattering parameters from real objects is still a challenging work. To obtain the scattering parameters, physics-based analysis is ineffective because huge computational cost is required to simulate subsurface scattering effect accurately. Thus, we focus on Curvature-Dependent Reflectance Function (CDRF), the plausible approximation of the subsurface scattering effect. In this paper, we propose a novel technique to obtain scattering parameters from real objects by revealing the relation between curvature and translucency.

© Springer International Publishing Switzerland 2016. This paper describes a method for freely changing the length of a video clip, leaving its content almost unchanged, by removing video frames considering both audio and video transitions. In a video clip that contains many video frames, there are less important frames that only extend the length of the clip. Taking the continuity of audio and video frames into account, the method enables changing the length of a video clip by removing or inserting frames that do not significantly affect the content. Our method can be used to change the length of a clip without changing the playback speed. Subjective experimental results demonstrate the effectiveness of our method in preserving the clip content.

© Springer International Publishing Switzerland 2016. This paper presents MusicMixer, an automatic DJ system that mixes songs in a seamless manner. MusicMixer mixes songs based on audio similarity calculated via beat analysis and latent topic analysis of the chromatic signal in the audio. The topic represents latent semantics about how chromatic sounds are generated. Given a list of songs, a DJ selects a song with beat and sounds similar to a specific point of the currently playing song to seamlessly transition between songs. By calculating the similarity of all existing pairs of songs, the proposed system can retrieve the best mixing point from innumerable possibilities. Although it is comparatively easy to calculate beat similarity from audio signals, it has been difficult to consider the semantics of songs as a human DJ considers. To consider such semantics, we propose a method to represent audio signals to construct topic models that acquire latent semantics of audio. The results of a subjective experiment demonstrate the effectiveness of the proposed latent semantic analysis method.

© Springer International Publishing Switzerland 2016. This paper presents Computational Cartoonist, a comicstyle anime summarization system that detects key frame and generates comic layout automatically. In contract to previous studies, we define evaluation criteria based on the correspondence between anime films and original comics to determine whether the result of comic-style summarization is relevant. To detect key frame detection for anime films, the proposed system segments the input video into a series of basic temporal units, and computes frame importance using image characteristics such as motion. Subsequently, comic-style layouts are decided on the basis of pre-defined templates stored in a database. Several results demonstrate the efficiency of our key frame detection over previous methods by evaluating the matching accuracy between key frames and original comic panels.

In this paper, we present a soundtrack generation system that can automatically add a soundtrack with the length and climax points aligned to those of a video clip. Adding a soundtrack to a video clip is an important process in video editing. Editors tend to add chorus sections to the climax points of the video clip by replacing and concatenating musical segments. However, this process is time-consuming Our system automatically detects climaxes of both the video clips and music based on feature extraction and analysis. This enables the system to add a soundtrack in which the climax is synchronized to the climax of the video clip. We evaluated the generated soundtracks through a subjective evaluation.

© 2016 The Eurographics Association. Subsurface scattering is important to express translucent materials such as skin, marble and so on realistically. However, rendering translucent materials in real-time is challenging, since calculating subsurface light transport requires large computational cost. In this paper, we present a novel algorithm to render heterogeneous translucent materials using Dijkstra's Algorithm. Our two main ideas are follows: The first is fast construction of the graph by solid voxelization. The second is voxel shading like initialization of the graph. From these ideas, we obtain maximum contribution of emitted light on whole surface in single calculation. We realize real-time rendering of animated heterogeneous translucent objects with simple compute and our approach does not require any precomputation.

© 2016 The Eurographics Association. Rendering of tranlucent objects enhances the reality of computer graphics, however, it is still computationally expensive. In this paper, we introduce a real-time rendering technique for heterogeneous translucent objects that contain complex structure inside. First, for the precomputation, we convert mesh models into voxels and generate Look-Up-Table in which the optical thickness between two surface voxels is stored. Second, we compute radiance in real-time using the precomputed optical thickness. At this time, we generate Voxel-Number-Map to fetch the texel value of the Look-Up-Table in GPU. Using Look-Up-Table and Voxel-Number-Map, our method can render translucent objects with cavities and different media inside in real-time.

An appearance of a human face changes due to aging: sagging, spots, lusters, and wrinkles would be observed. Therefore, facial aging simulation techniques are required for long-term criminal investigation. While the appearance of an aged face varies greatly from person to person, wrinkles are one of the most important features which represent the human individuality. An individuali-ty of wrinkles is defined by wrinkles shape and position. [Maejima et al. 2014] proposed an aging simulation method that preserves facial parts individuality using a patch-based facial image reconstruction. Since few wrinkles can be observed on an input young face, it is difficult to represent wrinkles appearance only by the reconstruction. Therefore, a statistical wrinkles aging pattern model is introduced to produce natural looking wrinkles by selecting appropriate patches in an age-specific patch database. However, the variation of the statistical wrinkles patterns model is too limited to represent wrinkles individuality. Additionally, an appropriate size of a patch and feature value had to be applied for each facial region to get a plausible aged facial image. In this paper, we introduce a novel aging simulation method us-ing the patch-based image reconstruction, which can overcome problems mentioned above. Based on a medical knowledge [Piér-ard et al. 2003], wrinkles in an expressive facial image (defined as expressive wrinkles) of a same person are synthesized to the input image instead of the statistical wrinkles pattern model to represent wrinkles individuality. Furthermore, different sizes of the patch and feature values are applied for each facial region to achieve high representation of both the wrinkles individuality and the age-specific features. The entire process is performed automatically, and a plausible aged facial image is generated.

In recent years, a lot of 3D character dance animation movies are created by amateur users using 3DCG animation editing tools (e.g. MikuMikuDance). Whereas, most of them are created manually. Then automatic facial animation system for dancing character will be useful to create dance movies and visualize impressions effec- tively. Therefore, we address the challenging theme to estimate dancing character's emotions (we call "dance emotion"). In previ- ous work considering music features, DiPaola et al. [2006] pro- posed music-driven emotionally expressive face system. To de- tect the mood of the input music, they used a hierarchical frame- work (Thayer model), and achieved to generate facial animation that matches music emotion. However, their model can't express subtleties of emotion between two emotions because input music divided into few moods sharply using Gaussian mixture model. In addition, they decide more detailed moods based on the psychologi- cal rules that uses score information, so they requires MIDI data. In this paper, we propose "dance emotion model" to visualize danc- ing character's emotion as facial expression. Our model is built by the coordinate information frame by frame on the emotional space through perceptional experiment using music and dance mo- tion database without MIDI data. Moreover, by considering the displacement on the emotional space, we can express not only a certain emotion but also subtleties of emotions. As the result, our system got a higher accuracy comparing with the previous work. We can create the facial expression result soon by inputting audio data and synchronized motion. It is shown the utility through the comparison with previous work in Figure 1.

© 2015 ACM. In this paper, we present a novel method for creating a painted im- Age from a photograph using an existing painting as a style source. The core idea is to identify the corresponding objects in the two images in order to select patches more appropriately. We automat- ically make a region correspondence between the painted source image and the target photograph by computing color and texture feature distances. Next, we conduct a patch-based synthesis that preserves the appropriate source and target features. Unlike previ- ous example-based approaches of painting style transfer, our results successfully reflect the features of the source images even if the in- put images have various colors and textures. Our method allows us to automatically render a new painted image preserving the features of the source image.

© 2015 The Author(s) Computer Graphics Forum. Due to the recent advancement of computer graphics hardware and software algorithms, deformable characters have become more and more popular in real-time applications such as computer games. While there are mature techniques to generate primary deformation from skeletal movement, simulating realistic and stable secondary deformation such as jiggling of fats remains challenging. On one hand, traditional volumetric approaches such as the finite element method require higher computational cost and are infeasible for limited hardware such as game consoles. On the other hand, while shape matching based simulations can produce plausible deformation in real-time, they suffer from a stiffness problem in which particles either show unrealistic deformation due to high gains, or cannot catch up with the body movement. In this paper, we propose a unified multi-layer lattice model to simulate the primary and secondary deformation of skeleton-driven characters. The core idea is to voxelize the input character mesh into multiple anatomical layers including the bone, muscle, fat and skin. Primary deformation is applied on the bone voxels with lattice-based skinning. The movement of these voxels is propagated to other voxel layers using lattice shape matching simulation, creating a natural secondary deformation. Our multi-layer lattice framework can produce simulation quality comparable to those from other volumetric approaches with a significantly smaller computational cost. It is best to be applied in real-time applications such as console games or interactive animation creation.

Due to the recent advancement of computer graphics hardware and software algorithms, deformable characters have become more and more popular in real-time applications such as computer games. While there are mature techniques to generate primary deformation from skeletal movement, simulating realistic and stable secondary deformation such as jiggling of fats remains challenging. On one hand, traditional volumetric approaches such as the finite element method require higher computational cost and are infeasible for limited hardware such as game consoles. On the other hand, while shape matching based simulations can produce plausible deformation in real-time, they suffer from a stiffness problem in which particles either show unrealistic deformation due to high gains, or cannot catch up with the body movement. In this paper, we propose a unified multi-layer lattice model to simulate the primary and secondary deformation of skeleton-driven characters. The core idea is to voxelize the input character mesh into multiple anatomical layers including the bone, muscle, fat and skin. Primary deformation is applied on the bone voxels with lattice-based skinning. The movement of these voxels is propagated to other voxel layers using lattice shape matching simulation, creating a natural secondary deformation. Our multi-layer lattice framework can produce simulation quality comparable to those from other volumetric approaches with a significantly smaller computational cost. It is best to be applied in real-time applications such as console games or interactive animation creation.

© 2015 Information Processing Society of Japan. In this paper, we propose a novel method to generate highly photorealistic three-dimensional (3D) inner mouth animation that is well-fitted to an original ready-made speech animation using only frontal captured images and small-size databases. The algorithms are composed of quasi-3D model reconstruction and motion control of teeth and the tongue, and final compositing of photorealistic speech animation synthesis tailored to the original. In general, producing a satisfactory photorealistic appearance of the inner mouth that is synchronized with mouth movement is a very complicated and time-consuming task. This is because the tongue and mouth are too flexible and delicate to be modeled with the large number of meshes required. Therefore, in some cases, this process is omitted or replaced with a very simple generic model. Our proposed method, on the other hand, can automatically generate 3D inner mouth appearances by improving photorealism with only three inputs: an original tailor-made lip-sync animation, a single image of the speaker’s teeth, and a syllabic decomposition of the desired speech. The key idea of our proposed method is to combine 3D reconstruction and simulation with two-dimensional (2D) image processing using only the above three inputs, as well as a tongue database and mouth database. The satisfactory performance of our proposed method is illustrated by the significant improvement in picture quality of several tailor-made animations to a degree nearly equivalent to that of camera-captured videos.

In this paper, we propose a novel method to generate highly photorealistic three-dimensional (3D) inner mouth animation that is well-fitted to an original ready-made speech animation using only frontal captured images and small-size databases. The algorithms are composed of quasi-3D model reconstruction and motion control of teeth and the tongue, and final compositing of photorealistic speech animation synthesis tailored to the original. In general, producing a satisfactory photorealistic appearance of the inner mouth that is synchronized with mouth movement is a very complicated and time-consuming task. This is because the tongue and mouth are too flexible and delicate to be modeled with the large number of meshes required. Therefore, in some cases, this process is omitted or replaced with a very simple generic model. Our proposed method, on the other hand, can automatically generate 3D inner mouth appearances by improving photorealism with only three inputs: an original tailor-made lip-sync animation, a single image of the speaker’s teeth, and a syllabic decomposition of the desired speech. The key idea of our proposed method is to combine 3D reconstruction and simulation with two-dimensional (2D) image processing using only the above three inputs, as well as a tongue database and mouth database. The satisfactory performance of our proposed method is illustrated by the significant improvement in picture quality of several tailor-made animations to a degree nearly equivalent to that of camera-captured videos.

A reconstruction of a human face shape from a single image is an important theme for criminal investigation such as recognition of suspected people from surveillance cameras with only a few frames. It is, however, still difficult to recover a face shape from a non-frontal face image. Method using shading cues on a face depends on the lighting circumstance and cannot be adapted to images in which shadows occurs, for example [Kemelmacher et al. 2011]. On the other hand, [Blanz et al. 2004] reconstructed a shape by 3D Morphable Model (3DMM) only with facial feature points. This method, however, requires the pose-wise correspondences of vertices in the model to feature points of input image because a face contour cannot be seen when the facial direction is not the front. In this paper, we propose a method which can reconstruct a facial shape from a non-frontal face image only with a single general correspondence table. Our method searches for the correspondences of points on a facial contour in the iterative reconstruction process, and makes the reconstruction simple and stable.

Dressing virtual characters is necessary for many applications, while modeling clothing is a significant bottleneck. Therefore, it has been proposed that the idea of Garment Transfer for transfer-ring clothing model from one character to another character [Brouet et al. 2012]. In recent years, this idea has been extended to be applicable between characters in various poses and shapes [Narita et al. 2014]. However, texture design of clothing is not preserved in their method since they deform the source clothing model to fit the target body (see Figure 1(a)(c)). We propose a novel method to transfer garment while preserving its texture design. First, we cut the transferred clothing mesh model along the seam. Second, we follow the similar method to "as-rigid-as-possible" deformation, we deform the texture space to reflect the shape of transferred clothing mesh model. Our method keeps consistency of the texture as clothing by cutting them along a seam. In order not to generate the inversion, we modify the ex-pression of "as-rigid-as-possible". Our method allows users not only to preserve texture uniformly on transferred clothing (see Figure 1(b)), but also in particular location the user specified, such as the location with an appliqué (see Figure 1(e)). Our meth-od is the pioneer of Texture Preserving Garment Transfer.

Anime designers often paint actual sceneries to serve as background images based on photographs to complement characters. As paint- ing background scenery is time consuming and cost ineffective, there is a high demand for techniques that can convert photographs into anime styled graphics. Previous approaches for this purpose, such as Image Quilting [Efros and Freeman 2001] transferred a source texture onto a target photograph. These methods synthesized corresponding source patches with the target elements in a photo- graph, and correspondence was achieved through nearest-neighbor search such as PatchMatch [Barnes et al. 2009]. However, the near- est-neighbor patch is not always the most suitable patch for anime transfer because photographs and anime background images differ in color and texture. For example, real-world color need to be con- verted into specific colors for anime; further, the type of brushwork required to realize an anime effect, is different for different photo- graph elements (e.g. sky, mountain, grass). Thus, to get the most suitable patch, we propose a method, wherein we establish global region correspondence before local patch match. In our proposed method, BGMaker, (1) we divide the real and anime images into re- gions; (2) then, we automatically acquire correspondence between each region on the basis of color and texture features, and (3) search and synthesize the most suitable patch within the corresponding re- gion. Our primary contribution in this paper is a method for au- Tomatically acquiring correspondence between target regions and source regions of different color and texture, which allows us to generate an anime background image while preserving the details of the source image.

In movie and video game productions, synthesizing subtle eye and corresponding head movements of CG character is essential to make a content dramatic and impressive. However, to complete them costs a lot of time and labors because they often have to be made by manual operations of skilled artists. [Itti et al. 2006] and [Yeo et al. 2012] proposed an automatic eyes and head's motion control method by measuring a real person watching a displayed gaze point. However, in both approaches, a rotational angle and speed of eyes and head are treated together uniformly depending on the gaze point location. Specifically, dis-playing duration time of gaze target strongly influences the motion of eyes and head because the shorter the blink interval of a gaze target is, the more quickly a human response becomes to chase the target by the combination of eye rotation and head movement. In this paper, we propose a method to automatically control eyes and head by taking account of both gaze target location and its blink time duration. As a result, eye and head movement are mod-eled combined with measured data by a function whose arguments are gaze point angle and duration time. So a variety of gaze action along with head motion including Ves-tibule-ocular Reflex can be generated automatically by changing the parameters of a gaze angle and duration.

This paper presents a system that can automatically add a sound-track to a video clip by replacing and concatenating an existing song's musical bars considering a user's preference. Since a soundtrack makes a video clip attractive, adding a soundtrack to a clip is one of the most important processes in video editing. To make a video clip more attractive, an editor of the clip tends to add a soundtrack considering its timing and climax. For example, editors often add chorus sections to the climax of the clip by re-placing and concatenating musical bars in an existing song. How-ever, in the process, editors should take naturalness of rearranged soundtrack into account. Therefore, editors have to decide how to replace musical bars in a song considering its timing, climax, and naturalness of rearranged soundtrack simultaneously. In this case, editors are required to optimize the soundtrack by listening to the rearranged result as well as checking the naturalness and synchro-nization between the result and the video clip. However, this repe-titious work is time-consuming. [Feng et al. 2010] proposed an automatic soundtrack addition method. However, since this meth-od automatically adds soundtrack with data-driven approach, this method cannot consider timing and climax which a user prefers. Our system takes all the patterns of rearranged musical bars into account and finds the most natural soundtrack considering a user's preference of intention for an audio-visual alignment and a climax of the resulting soundtrack. Specifically, musical sections between user's specified points and the beginning and the ending of the song are automatically interpolated by replacing and concatenat-ing musical bars based on dynamic programming. To consider user's intention for a climax of the soundtrack, the system allows the user to specify the intended climax by an editing interface. The system immediately reflects the intention and the soundtrack will be interactively re-rearranged. These semi-automated pro-cesses of rearranging soundtrack for a video clip help the users adding songs without considering naturalness of the rearranged song by their own.

This paper presents a method to synthesize a high speed soft body character animation in real time. Especially, not only primary but also secondary physics based deformation in an under skin fat structure driven by a skeletal motion are considered. However, as a high fidelity soft body animation method, high cost FEM based methods are usually introduced, the method which represents a high speed and robust soft body animation using simplified elastic simulation method was proposed. In this method, they applied skinning result to fat and skin, and then a vibration was limited to be a very small and unnatural. So in this paper, we proposed a method to divide skinning and simulation by modeling an inside structure under skin approximately and automatically. As a result, by controlling parameters such as volume and elastic parameter at each layer freely. Consequently, by a user interaction with parameter adjustment, a definition of soft body material according to the location can be implemented to generate and edit a proper soft body motion. After an evaluation of soft body motion synthesis based on a several character models, this method is approved to be very effective to make a character animation with soft body characters.

The topic of face-aging received increased attention by the computer vision community during the recent years. This interest is motivated by important real life applications where accurate age progression algorithms can be used. However, age progression methodologies may only be used in real applications provided that they have the ability to produce accurate age progressed images. Therefore it is of utmost importance to encourage the development of accurate age progression algorithms through the formulation of performance evaluation protocols that can be used for obtaining accurate performance evaluation results for different algorithms reported in the literature. In this paper we describe the organization of the, first ever, pilot independent age progression competition that aims to provide the basis of a robust framework for assessing age progression methodologies. The evaluation carried out involves the use of several machine-based and human-based indicators that were used for assessing eight age progression methods.

The topic of face-aging received increased attention by the computer vision community during the recent years. This interest is motivated by important real life applications where accurate age progression algorithms can be used. However, age progression methodologies may only be used in real applications provided that they have the ability to produce accurate age progressed images. Therefore it is of utmost importance to encourage the development of accurate age progression algorithms through the formulation of performance evaluation protocols that can be used for obtaining accurate performance evaluation results for different algorithms reported in the literature. In this paper we describe the organization of the, first ever, pilot independent age progression competition that aims to provide the basis of a robust framework for assessing age progression methodologies. The evaluation carried out involves the use of several machine-based and human-based indicators that were used for assessing eight age progression methods.

The topic of face-aging received increased attention by the computer vision community during the recent years. This interest is motivated by important real life applications where accurate age progression algorithms can be used. However, age progression methodologies may only be used in real applications provided that they have the ability to produce accurate age progressed images. Therefore it is of utmost importance to encourage the development of accurate age progression algorithms through the formulation of performance evaluation protocols that can be used for obtaining accurate performance evaluation results for different algorithms reported in the literature. In this paper we describe the organization of the, first ever, pilot independent age progression competition that aims to provide the basis of a robust framework for assessing age progression methodologies. The evaluation carried out involves the use of several machine-based and human-based indicators that were used for assessing eight age progression methods.

The topic of face-aging received increased attention by the computer vision community during the recent years. This interest is motivated by important real life applications where accurate age progression algorithms can be used. However, age progression methodologies may only be used in real applications provided that they have the ability to produce accurate age progressed images. Therefore it is of utmost importance to encourage the development of accurate age progression algorithms through the formulation of performance evaluation protocols that can be used for obtaining accurate performance evaluation results for different algorithms reported in the literature. In this paper we describe the organization of the, first ever, pilot independent age progression competition that aims to provide the basis of a robust framework for assessing age progression methodologies. The evaluation carried out involves the use of several machine-based and human-based indicators that were used for assessing eight age progression methods.

© Springer International Publishing Switzerland 2015. We present an affective music recommendation system just fitting to an input video without textual information. Music that matches our current environmental mood can enhance a deep impression. However, we cannot know easily which music best matches our present mood from huge music database. So we often select a well-known popular song repeatedly in spite of the present mood. In this paper, we analyze the video sequence which represent current mood and recommend an appropriate music which affects the current mood. Our system matches an input video with music using valence-arousal plane which is an emotional plane.

© Springer International Publishing Switzerland 2015. Facial aging and rejuvenation simulation is a challenging topic because keeping personal characteristics in every age is difficult problem. In this demonstration, we simulate a facial aging/rejuvenating only from a single photo. Our system alters an input face image to aged face by reconstructing every facial component with face database for target age. An appropriate facial components image are selected by a special similarity measurement between current age and target age to keep personal characteristics asmuch as possible. Our systemsuccessfully generated aged/ rejuvenated faces with age-related features such as spots, wrinkles, and sagging while keeping personal characteristics throughout all ages.

© Springer International Publishing Switzerland 2015. Facial image synthesis creates blurred facial images almost without high-frequency components, resulting in flat edges. Moreover, the synthesis process results in inconsistent facial images, such as the conditions where the white part of the eye is tinged with the color of the iris and the nasal cavity is tinged with the skin color. Therefore, we propose a method that can deblur an inconsistent synthesized facial image, including strong blurs created by common image morphing methods, and synthesize photographic quality facial images as clear as an image captured by a camera. Our system uses two original algorithms: patch color transfer and patch-optimized visio-lization. Patch color transfer can normalize facial luminance values with high precision, and patch-optimized visio-lization can synthesize a deblurred, photographic quality facial image. The advantages of our method are that it enables the reconstruction of the high-frequency components (concavo-convex) of human skin and removes strong blurs by employing only the input images used for original image morphing.

In this paper, we propose a novel facial fattening and slimming deformation method in 2D images that preserves the individuality of the input face by estimating the skull structure from a frontal face image and prevents unnatural deformation (e.g. penetration into the skull). Our method is composed of skull estimation, optimizing fattening and slimming rules appropriate to the estimated skull, mesh deformation to generate fattening and slimming face, and generation background image adapted to the generated face contour. Finally, we verify our method by comparison with other rules, precision of skull estimation, subjective experiment, and execution time.

Copyright © 2015 SCITEPRESS - Science and Technology Publications. All rights reserved. Data-driven animation using a large human motion database enables the programing of various natural human motions. While the development of a motion capture system allows the acquisition of realistic human motion, segmenting the captured motion into a series of primitive motions for the construction of a motion database is necessary. Although most segmentation methods have focused on periodic motion, e.g., walking and jogging, segmenting non-periodic and asymmetrical motions such as dance performance, remains a challenging problem. In this paper, we present a specialized segmentation approach for human dance motion. Our approach consists of three steps based on the assumption that human dance motion is composed of consecutive choreographic primitives. First, we perform an investigation based on dancer perception to determine segmentation components. After professional dancers have selected segmentation sequences, we use their selected sequences to define rules for the segmentation of choreographic primitives. Finally, the accuracy of our approach is verified by a user-study, and we thereby show that our approach is superior to existing segmentation methods. Through three steps, we demonstrate automatic dance motion synthesis based on the choreographic primitives obtained.

The topic of face-aging received increased attention by the computer vision community during the recent years. This interest is motivated by important real life applications where accurate age progression algorithms can be used. However, age progression methodologies may only be used in real applications provided that they have the ability to produce accurate age progressed images. Therefore it is of utmost importance to encourage the development of accurate age progression algorithms through the formulation of performance evaluation protocols that can be used for obtaining accurate performance evaluation results for different algorithms reported in the literature. In this paper we describe the organization of the, first ever, pilot independent age progression competition that aims to provide the basis of a robust framework for assessing age progression methodologies. The evaluation carried out involves the use of several machine-based and human-based indicators that were used for assessing eight age progression methods.

In this paper, we present MusicMixer, a computer-aided DJ system that helps DJs, specifically with song mixing. MusicMixer continuously mixes and plays songs using an automatic music mixing method that employs audio similarity calculations. By calculating similarities between song sections that can be naturally mixed, MusicMixer enables seamless song transitions. Though song mixing is the most fundamental and important factor in DJ performance, it is difficult for untrained people to seamlessly connect songs. MusicMixer realizes automatic song mixing using an audio signal processing approach; therefore, users can perform DJ mixing simply by selecting a song from a list of songs suggested by the system, enabling effective DJ song mixing and lowering entry barriers for the inexperienced. We also propose personalization for song suggestions using a preference memorization function of MusicMixer.

Copyright © 2014 ACM. This paper presents VRMixer, a system that mixes real world and a video clip letting a user enter the video clip and realize a virtual co-starring role with people appearing in the clip. Our system constructs a simple virtual space by allocating video frames and the people appearing in the clip within the user's 3D space. By measuring the user's 3D depth in real time, the time space of the video clip and the user's 3D space become mixed. VRMixer automatically extracts human images from a video clip by using a video segmentation technique based on 3D graph cut segmentation that employs face detection to detach the human area from the background. A virtual 3D space (i.e., 2.5D space) is constructed by positioning the background in the back and the people in the front. In the video clip, the user can stand in front of or behind the people by using a depth camera. Real objects that are closer than the distance of the clip's background will become part of the constructed virtual 3D space. This synthesis creates a new image in which the user appears to be a part of the video clip, or in which people in the clip appear to enter the real world. We aim to realize "video reality," i.e., a mixture of reality and video clips using VRMixer.

Copyright © ACM. This paper presents a method that enables the estimation and depiction of onomatopoeia in computer-generated animation based on physical parameters. Onomatopoeia is used to enhance physical characteristics and movement, and enables users to understand animation more intuitively. We experiment with onomatopoeia depiction in scenes within the animation process. To quantify onomatopoeia, we employ Komatsu's [2012] assumption, i.e., onomatopoeia can be expressed by n-dimensional vector. We also propose phonetic symbol vectors based on the correspondence of phonetic symbols to the impressions of onomatopoeia using a questionnaire-based investigation. Furthermore, we verify the positioning of onomatopoeia in animated scenes. The algorithms directly combine phonetic symbols to estimate optimum onomatopoeia. They use a view-dependent Gaussian function to display onomatopoeias in animated scenes. Our method successfully recommends optimum onomatopoeias using only physical parameters, so that even amateur animators can easily create onomatopoeia animation.

本稿では，ライヴ動画や PV などに代表される音楽動画において，歌手が歌っているシーンである 「歌唱シーン」 を検出する手法について検討する．音楽において歌手は最も主要な役割を担っており，音楽動画における歌唱シーンも同様に動画のハイライトの一つであると言える．歌唱シーンは動画サムネイル生成や，大量の音楽動画の短時間ブラウジングなどにおいて有用である．歌唱シーンを検出するためには，歌手の顔認識，楽曲中の歌声区間検出といった要素手法及びそれらを組み合わせる手法についての検討が必要である．本稿では，顔認識を用いた映像解析，歌声区間検出を用いた音響解析，それらを複合した Audio-visual 解析のそれぞれについて比較・検討しながら歌唱シーン検出の実現可能性について議論する．

Various methods of predicting the deformation of fine-scale cloth from coarser resolutions have been explored. However, the influence of fine-scale deformation has not been considered in coarse-scale simulations. Thus, the simulation of highly nonhomogeneous detailed cloth is prone to large errors. We introduce an effective method to simulate cloth made of nonhomogeneous, anisotropic materials. We precompute a macroscopic stiffness that incorporates anisotropy from the microscopic structure, using the deformation computed for each unit strain. At every time step of the simulation, we compute the deformation of coarse meshes using the coarsened stiffness, which saves computational time and add higher-level details constructed by the characteristic displacement of simulated meshes. We demonstrate that anisotropic and inhomogeneous cloth models can be simulated efficiently using our method. (c) 2014 The Authors. Computer Animation and Virtual Worlds published by John Wiley & Sons, Ltd.

We propose a method for estimating the degree of a driver's drowsiness on the basis of changes in facial expressions captured by an IR camera. Typically, drowsiness is accompanied by drooping eyelids. Therefore, most related studies have focused on tracking eyelid movement by monitoring facial feature points. However, the drowsiness feature emerges not only in eyelid movements but also in other facial expressions. To more precisely estimate drowsiness, we must select other effective features. In this study, we detected a new drowsiness feature by comparing a video image and CG model that are applied to the existing feature point information. In addition, we propose a more precise degree of drowsiness estimation method using wrinkle changes and calculating local edge intensity on faces, which expresses drowsiness more directly in the initial stage.

In this paper, we present a novel approach that utilizes the geometry shader to dynamically voxelize 3D models in real-time. In the geometry shader, the primitives are split by their Z-order, and then rendered to tiles which compose a single 2D texture. This method is completely based on graphic pipeline, rather than computational methods like CUDA/OpenCL implementation. So it can be easily integrated into a rendering or simulation system. Another advantage of our algorithm is that while doing voxelization, it can simultaneously record the additional mesh information like normal, material properties and even speed of vertex displacement. Our method achieves conservative voxelization by only two passes of rendering without any preprocessing and it fully runs on GPU. As a result, our algorithm is very useful for dynamic application. © 2014 Springer International Publishing.

© Taro Masuda, Kazuyoshi Yoshii, Masataka Goto, Shigeo Morishima. This paper proposes a query-by-audio system that aims to detect temporal locations where a musical phrase given as a query is played in musical pieces. The “phrase” in this paper means a short audio excerpt that is not limited to a main melody (singing part) and is usually played by a single musical instrument. A main problem of this task is that the query is often buried in mixture signals consisting of various instruments. To solve this problem, we propose a method that can appropriately calculate the distance between a query and partial components of a musical piece. More specifically, gamma process nonnegative matrix factorization (GaP-NMF) is used for decomposing the spectrogram of the query into an appropriate number of basis spectra and their activation patterns. Semi-supervised GaP-NMF is then used for estimating activation patterns of the learned basis spectra in the musical piece by presuming the piece to partially consist of those spectra. This enables distance calculation based on activation patterns. The experimental results showed that our method outperformed conventional matching methods.

Speech animation synthesis is still a challenging topic in the field of computer graphics. Despite many challenges, representing detailed appearance of inner mouth such as nipping tongue's tip with teeth and tongue's back hasn't been achieved in the resulting animation. To solve this problem, we propose a method of data-driven speech animation synthesis especially when focusing on the inside of the mouth. First, we classify inner mouth into teeth labeling opening distance of the teeth and a tongue according to phoneme information. We then insert them into existing speech animation based on opening distance of the teeth and phoneme information. Finally, we apply patch-based texture synthesis technique with a 2,213 images database created from 7 subjects to the resulting animation. By using the proposed method, we can automatically generate a speech animation with the realistic inner mouth from the existing speech animation created by previous methods. © 2014 Information Processing Society of Japan.

Speech animation synthesis is still a challenging topic in the field of computer graphics. Despite many challenges, representing detailed appearance of inner mouth such as nipping tongue's tip with teeth and tongue's back hasn't been achieved in the resulting animation. To solve this problem, we propose a method of data-driven speech animation synthesis especially when focusing on the inside of the mouth. First, we classify inner mouth into teeth labeling opening distance of the teeth and a tongue according to phoneme information. We then insert them into existing speech animation based on opening distance of the teeth and phoneme information. Finally, we apply patch-based texture synthesis technique with a 2,213 images database created from 7 subjects to the resulting animation. By using the proposed method, we can automatically generate a speech animation with the realistic inner mouth from the existing speech animation created by previous methods. © 2014 Information Processing Society of Japan.

© Springer International Publishing Switzerland 2014. In this paper, we propose a novel method to generate highly photorealistic three-dimensional (3D) inner mouth animation that is well-fitted to an original ready-made speech animation using only frontal captured images and a small-size database. The algorithms are composed of quasi-3D model reconstruction and motion control of teeth and the tongue, and final compositing of photorealistic speech animation synthesis tailored to the original.

It is important to generate realistic eye movements by Computer Graphics. However, previous methods require a lot of expense and work when we generate eye movements in conversations. In this paper, we propose a method which automatically generates eye movements during conversation based on probability models derived from actual measurements. First, we measure eye movements with blinks in conversations and fixation eye movements. Second, we classify eye movements into saccades and fixation eye movements based on measurement result. Third, we approximate saccades, fixation eye movements and blinks using probability models and apply them to characters. As a result, we can generate realistic eye movements automatically.

Recently, in purpose of video summarization, methods of comic-like video summarization by play-log or image features are proposed. In many previous methods, there is no mention about amount and granularity of information to summarize video properly. This is because there is evaluation scale to decide whether the key-frames which obtained by these previous methods are appropriate as a comic. Therefore, we propose a scale to evaluate video summarization by using matching rate of animation and original comic. In addition, we propose method to decide key-frames which summarize video properly as a comic by image features. By calculating importance of both shot and frame after segmenting video into each scenes, we improved appropriateness of key-frames as a comic.

本研究では，ある動画のシーンに類似した別の動画のシーンを検索するための映像特徴量間の類似尺度を提案する。類似動画検索に関する多くの研究は，特徴量の計算を画面全体に対して行っているため，画面の中で重要な部分とそうでない部分を等しく扱っている。そこで本手法では，映像において人の注意を引き付けやすい領域を推定し，その局所領域内での特徴を抽出する。これに画面全体の特徴も加味することで，大局的な特徴と局所的な特徴の両方を考慮した動画間の類似尺度を構成する。クエリとなる入力動画のシーンと最も類似度の高い動画シーンをデータベースから探索することで類似動画の検索を行った。また，実験により本手法の有効性についても検討した。

Simulation of the reflectance of translucent materials is one of the most important factors in the creation of realistic CG objects. Estimating the reflectance characteristics of translucent materials from a single image is a very efficient way of re-rendering objects that exist in real environments. However, this task is considerably challenging because this approach leads to problems such as the existence of many unknown parameters. Munoz et al. [2011] proposed a method for the estimation of the bidirectional surface scattering reflectance distribution function (BSSRDF) from a given single image. However, it is difficult or impossible to estimate the BSSRDF of materials with complex shapes because this method's target was the convexity of objects therefore, it used a rough depth recovery technique for global convex objects. In this paper, we propose a method for accurately estimating the BSSRDF of human faces, which have complex shapes. We use a 3D face reconstruction technique to satisfy the above assumption. We are able to acquire more accurate geometries of human faces, and it enables us to estimate the reflectance characteristics of faces.

When we synthesize a realistic appearance of dust-covered object by CG, it is necessary to express a large number of fabric components of dust accurately with many short fibers, and as a result, this process is a time-consuming task. The dust amount prediction function suggested by Hsu [1995] proposed modeling and rendering techniques for dusty surfaces. These techniques only describe dust accumulation as a shading function, however, they cannot express the volume of dust on the surfaces. In this study, we present a novel method to model and render the appearance and volume of dust in real-time by using shell texturing. Each shell texture, which can express several components, is automatically generated in our procedural approach. Therefore, we can draw any arbitrary appearance of dust rapidly and interactively by solely controlling simple parameters.

Music that matches our current mood can create a deep impression, which we usually want to enjoy when we listen to music. However, we do not know which music best matches our present mood. We have to listen to each song, searching for music that matches our mood. As it is difficult to select music manually, we need a recommendation system that can operate affectively. Most recommendation methods, such as collaborative filtering or content similarity, do not target a specific mood. In addition, there may be no word exactly specifying the mood. Therefore, textual retrieval is not effective. In this paper, we assume that there exists a relationship between our mood and images because visual information affects our mood when we listen to music. We now present an affective music recommendation system using an input image without textual information.

Many studies on an aged face image synthesis have been reported with the purpose of security application such as investigation for criminal or kidnapped child and entertainment applications such as movie or video game.

In recent years, the rate of fatal motor vehicle accidents caused by distracted driving resulting from factors such as sleeping at the wheel has been increasing. Therefore, an alert system that detects driver drowsiness and prevents accidents as a result by warning drivers before they fall asleep is urgently required. Non-contact measuring systems using computer vision techniques have been studied, and in vision approach, it is important to decide what kind of feature we should use for estimating drowsiness.

We often see close-ups of CG characters' faces in movies or video games. In such situations, the quality of a character's face (mainly in dialogue scenes) primarily determines that of the entire movie. Creating highly realistic speech animation is essential because viewers watch these scenes carefully. In general, such speech animations are created manually by skilled artists. However, creating them requires a considerable effort and time.

Generating realistic eye movements is a significant topic in Computer Graphics(CG) contents production field. Appropriate modeling and synthesis for eye movements are greatly difficult because they have a lot of important features. Gu et al[2007] proposed a method for automatically synthesizing realistic eye movements during conversations according to probability models. Despite eye movements during conversations include both saccades and fixational eye movements (FEMs), they synthesized only saccades which are relatively large eye movements.

The power of expression such as accent in motion and movement of arms is an indispensable factor in dance performance because there is a large difference in appearance between natural dance and expressive motions. Needless to say, expressive dance motion makes a great impression on viewers. However, creating such a dance motion is challenging because most of the creators have little knowledge about dance performance. Therefore, there is a demand for a system that generates expressive dance motion with ease. Tsuruta et al. [2010] generated expressive dance motion by changing only the speed of input motion or altering joint angles. However, the power of expression was not evaluated with certainty, and the generated motion did not synchronize with music. Therefore, the generated motion did not always satisfy the viewers.

Computer-generated speech animations are commonly seen in video games and movies. Although high-quality facial motions can be created by the hand crafted work of skilled artists, this approach is not always suitable because of time and cost constraints. A data-driven approach [Taylor et al. 2012], such as machine learning to concatenate video portions of speech training data, has been utilized to generate natural speech animation, while a large number of target shapes are often required for synthesis. We can obtain smooth mouth motions from prepared lip shapes for typical vowels by using an interpolation of lip shapes with Gaussian mixture models (GMMs) [Yano et al. 2007]. However, the resulting animation is not directly generated from the measured lip motions of someone's actual speech.

Hair rendering and simulation is a fundamental part in the representation of virtual characters. But intensive calculation for the dynamic on thousands of hair strands makes the task much challengeable, especially on a portable device. The aim of this short paper is to solve the problem of how to perform real-time hair simulation and rendering on mobile device. In this paper, the process of hair simulation and rendering is adapted according to the property of mobile device hardware. To increase the number of hair strands of simulation, we adopted the Dynamic follow-the-leader (DFTL) method and altered it by our new method of interpolation. We also pictured a rendering strategy basing on the survey of the limitation of mobile GPU. Lastly we present an innovational method that carried out order independent transparency at a relatively inexpensive cost. © 2013 ACM.

Music video is a short film which presents a visual representation of recent music. In these days, there is a trend that amateur users create music video in the video sharing website. Especially, the music video which is created by cutting and pasting existing video is called mashup music video. In this paper, we proposed the system that users can easily create mushup music video by using existing music videos. In addition, we conducted assessment evaluation experiment for our system. The system firstly extracts music features and video features from existing music videos. Then, the each feature is clustered and the relationship between each feature is learned by Hidden Markov Model. At last, the system cuts learned video scene which is the closest feature among learned videos and pastes it synchronizing with input song. Experiment shows that our method can generate more synchronized video than a previous method.

We present an affective music recommendation system using input images without textual information. Music that matches our current mood can create a deep impression. However, we do not know which music best matches our present mood. As it is difficult to select music manually, we need a recommendation system that can operate affectively. In this paper, we assume that there exists a relationship between our mood and images because visual information affects our mood when we listen to music. Our system matches an input image with music using valence-arousal plane which is an emotional plane.

Recently, many studies on facial aging synthesis have been reported for the purpose of security applications such as criminal investigation, kidnappings, and entertainment applications such as movies or video games. However, the representation of wrinkles, which is one of the most important elements when reflecting age characteristics, remains difficult. Additionally, the influence of lighting conditions and every individual's skin color is significant, and it is difficult to infer the location and shape of future wrinkles because they depend on factors such as one's living environment, eating habits, and DNA. Therefore, we must consider several possibilities for the locations of wrinkles. In this paper, we propose a facial aging synthesis method that can create plausible aged facial images, and is able to represent wrinkles at any desired location by drawing artificial freehand wrinkles while retaining photoreafistic quality.

We propose a method for the estimation of the degree of a driver's drowsiness on basis of changes in facial expressions captured by an IR camera. Typically, drowsiness is accompanied by falling of eyelids. Therefore, most of the related studies have focused on tracking eyelid movement by monitoring facial feature points. However, textural changes that arise from frowning are also very important and sensitive features in the initial stage of drowsiness, and it is difficult to detect such changes solely using facial feature points. In this paper, we propose a more precise drowsiness-degree estimation method considering wrinkles change by calculating local edge intensity on faces that expresses drowsiness more directly in the initial stage.

We propose a facial aging simulator which can synthesize a photorealistic human aged-face image for criminal investigation. Our aging simulator is based on the patch-based facial texture reconstruction with a wrinkle aging pattern model. The advantage of our method is to synthesize an aged-face image with detailed skin texture such as spots and somberness of facial skin, as well as age-related facial wrinkles without blurs that are derived from lack of accurate pixel-wise alignments as in the linear combination model, while maintaining the identity of the original face.

People can estimate an approximate age of others by looking at their faces. This is because faces have certain elements by which people can judge a person's age. If computers can extract and manipulate such information, wide variety of applications for entertainment and security purpose would be expected. © 2012 ACM.

We can visualize acquaintances' appearance by just hearing their voice if we have met them in past few years. Thus, it would appear that some relationships exist in between voice and appearance. If 3D head geometry could be estimated from a voice, we can realize some applications (e.g, avatar generation, character modeling for video game, etc.). Previously, although many researchers have been reported about a relationship between acoustic features of a voice and its corresponding dynamical visual features including lip, tongue, and jaw movements or vocal articulation during a speech, however, there have been few reports about a relationship between acoustic features and static 3D head geometry. In this paper, we focus on estimating 3D head geometry from a voice. Acoustic features vary depending on a speech context and its intonation. Therefore we restrict a context to Japanese 5 vowels. Under this assumption, to estimate 3D head geometry, we use a Feedforward Neural Network (FNN) trained by using a correspondence between an individual acoustic features extracted from a Japanese vowel and 3D head geometry generated based on a 3D range scan. The performance of our method is shown by both closed and open tests. As a result, we found that 3D head geometry which is acoustically similar to an input voice could be estimated under the limited condition. © 2012 ACM.

We present a real-time rendering method of translucent materials with complex shape by estimating object's thickness between light source and view point precisely. Wang et al. [2010] has already proposed a real-time rendering method treating arbitrary shapes, but it requires such huge computational costs and graphics memories that it is very difficult to implement in a practical rendering pipe-line. Inside a trans-lucent object, the energy of incident light attenuates highly depends on the object's optical thickness. Translucent Shadow Maps (TSM) [2003] is able to compute object's thickness using depth map at light position. However, TSM is not able to calculate thickness accurately in concave objects. In this paper, we propose a novel technique to compute object's thickness precisely and as a result, we achieve a real-time rendering of translucent materials with complex shapes only by adding one render-ing pass to conventional TSM. Copyright is held by the author / owner(s).

In this paper, we present a new method to generate a 3D face model, based on both Data-Driven and Structure-from-Motion approach. Considering both 2D frontal face image constraint, 3D geometric constraint, and likelihood constraint, we are able to reconstruct subject's face model accurately, robustly, and automatically. Using our method, it is possible to create a 3D face model in 5.8 [sec] by only shaking own head freely in front of a single video camera.

In this paper, we propose a method to apply any new face to a retro-projected 3D face system, the Mask-bot, which we have developed as a human-robot interface. The robot face using facial animation projected onto a 3D face mask can be quickly replaced by a new face based on a single frontal image of any person. Our contribution is to apply an automatic face replacement technique with the modified texture morphable model fitting to the 3D face mask. Using our technique, a face model displayed on Mask-bot can be automatically replaced within approximately 3 seconds, which makes Mask-bot widely suitable to applications such as video conferencing and cognitive experiments.

This paper presents an overview of the new version of our multi-model communication face "Mask-Bot", a rear-projected animated robotic head, including our display system, face animation, speech communication and sound localization.

In character animation field, many deformation techniques have been proposed. Example-based deformation methods are widely used especially for interactive applications. Example-based methods are mainly divided into two types. One is Interpolation. Methods in this type are designed to interpolate examples in a pose space. The advantage is that the deformed meshes can precisely correspond to the example meshes. On the other hand, the disadvantage is that larger number of examples is needed to generate arbitrary plausible interpolated meshes between each example. The other is Example-based Skinning which optimizes particular parameters referencing examples to represent example meshes as accurately as possible. These methods provide plausible deformations with fewer examples. However they cannot perfectly depict example meshes. In this paper, we present an idea that combines techniques belonging to the two types, taking advantages of both types. We propose an example-based skinning method to be combined with Pose Space Deformation (PSD). It optimizes transformation matrices in Skeleton Subspace deformation (SSD) introducing "support joints". Our method itself generates plausible intermediate meshes with a small set of examples as well as other example-based skinning methods. Then we explain the benefit of combining our method with PSD. We show that provided examples are precisely represented and plausible deformations at arbitrary poses are obtained by our integrated method.

To synthesize a realistic human animation using computer graphics, it is necessary to simulate subsurface scattering inside a human skin. We have developed a curvature-dependent reflectance functions (CDRF) which mimics the presence of a subsurface scattering effect. In this approach, we provide only a single parameter that represents the intensity of incident light scattering in a translucent material. We implemented our algorithm as a hardware-accelerated real-time renderer with a HLSL pixel shader. This approach is easily implementable on the GPU and does not require any complicated pre-processing and multi-pass rendering as is often the case in this area of research.

This paper presents a method to measure online the gait features from the gait silhouette images and to synthesize characteristic gait animation for an audience-participant digital entertainment. First, both static and dynamic gait features are extracted from the silhouette images captured by an online gait measurement system. Then, key motion data for various gaits are captured and a new motion data is synthesized by blending key motion data. Finally, blend ratios of the key motion data are estimated to minimize gait feature errors between the blended model and the online measurement. In experiments, the effectiveness of gait feature extraction were confirmed by using 100 subjects from OU-ISIR Gait Database and characteristic gait animations were created based on the measured gait features. © 2011 Springer-Verlag.

We propose a technique for automatically generating a realistic facial animation with precise individual facial geometry and characteristic facial expressions. Our method is divided into two key methods: the head modeling process automatically generates a whole head model only from facial range scan data, the facial animation setup process automatically generates key shapes which represent individual facial expressions based on physics-based facial muscle simulation with an individual muscle layout estimated from facial expression videos. Facial animations considering individual characteristics can be synthesized using the generated head model and key shapes. Experimental results show that the proposed method can generate facial animations where 84% of subjects can identify themselves. Therefore, we conclude that our head modeling techniques are effective to entertainment system like a Future Cast. © 2011 Springer-Verlag.

"Dive into the Movie (DIM)" is a name of project to aim to realize a world innovative entertainment system which can provide an immersion experience into the story by giving a chance to audience to share an impression with his family or friends by watching a movie in which all audience can participate in the story as movie casts. To realize this system, we are trying to model and capture the personal characteristics instantly and precisely in face, body, gait, hair and voice. All of the modeling, character synthesis, rendering and compositing processes have to be performed on real-time without any manual operation. In this paper, a novel entertainment system, Future Cast System (FCS), is introduced as a prototype of DIM. The first experimental trial demonstration of FCS was performed at the World Exposition 2005 in which 1,630,000 people have experienced this event during 6 months. And finally up-to-date DIM system to realize more realistic sensation is introduced. © 2011 Springer-Verlag.

The paper describes voice assignment techniques for synchronized scenario speech output in an instant casting movie system that enables anyone to be a movie star using his or her own voice and face. Two prototype systems were implemented, and both systems worked well for various participants, ranging from children to the elderly. © 2011 Springer-Verlag.

In this paper, we first present a new audience-participating movie form in which 3D virtual characters of audiences are constructed by computer graphics (CC) technologies and are embedded into a in a pre-rendered movie as different roles. Then, we investigate how the audiences respond to these virtual characters using physiological and subjective evaluation methods. To facilitate the investigation, we present three versions of a movie to an audience a Traditional version, its SDIM version with the participation of the audience's virtual character, and its SFDIM version with the co-participation of the audience and her/his friends' virtual characters. The subjective evaluation results show that the participation of virtual characters indeed causes increased subjective sense of spatial presence and engagement, and emotional reaction; moreover, SFDIM performs significantly better than SDIM, due to the co-participation of friends' virtual characters. Also, we find that the audiences experience not only significantly different galvanic skin response (GSR) changes on average changing trend over time and number of fluctuations but they also show the increased phasic GSR responses to the appearance of their own or friends' virtual 3D characters on the screen. The evaluation results demonstrate the success of the new audience-participating movie form and contribute to understanding how people respond to virtual characters in a role-playing entertainment interface. (C) 2009 Elsevier B.V. All rights reserved.

We propose a new automatic head modeling system to generate individualized head models which can express person-specific facial expressions. The head modeling system consists of two core processes. The head modeling process with the proposed automatic mesh completion generates a whole head model only from facial range scan data. The key shape generation process generates key shapes for the generated head model based on physics-based facial muscle simulation with an individual muscle layout estimated from subject's facial expression videos. Facial animations considering personal characteristics can be synthesized using the individualized head model and key shapes. Experimental results show that the proposed system can generate head models where 84% of subjects can identify themselves. Therefore, we conclude that our head modeling system is effective to games and entertainment systems like a Future Cast System. Copyright © 2010 by the Association for Computing Machinery, Inc.

Simulating sub-surface scattering is one of the most effective ways for realistically synthesizing translucent materials such as marble, milk and human skin. In previous work, the method developed by Jensen et al. [2002] significantly improved the speed of the simulation. However, the process is still not fast enough to produce realtime rendering. Thus, we have developed a curvature-dependent reflectance function (CDRF) which mimics the presence of a subsurface scattering effect.

This paper discusses an approach for computing the ambient occlusion by curvature depended approximation of occlusion. Ambient occlusion is widely used to improve the realism of fast lighting simulation. The ambient occlusion is defined as follows. © ACM 2010.

Realistic drape and motion of virtual clothing is now possible by using an up-to-date cloth simulator, but it is even difficult and time consuming to adjust and tune many parameters to achieve an authentic looking of a real particular fabric. Bhat et al. [2003] proposed a way to estimate the parameters from the video data of real fabrics. However, this projects structured light patterns on the fabrics, so it might not be possible to estimate the accurate value of the parameters if fabrics have colors and textures. In addition to the structured light patterns, they use a motion capture system to track how the fabrics move. In this paper, we will introduce a new method using only a motion capture system by attaching a few markers on fabric surface without any other devices. Moreover, animators can easily estimate the parameters of many kinds of fabrics with this method. Authentic looking and motion of simulated fabrics are realized by minimizing error function between captured motion data and synthetic motion considering both static and dynamic cloth features. © ACM 2010.

In anime production, some key-frames are drawn by artist precisely and then a great number of in-betweening frames are drawn by assistants' hands. However, it is seriously time-consuming and skilled work to draw many characters especially including face rotation. In this paper, we propose an automatic in-betweening technique for rotating face of hand drawn character only from a front image and a diagonal image (Fig. 1). Baxter [2009] represented generating in-betweening using image morphing technique. However, their approach doesn't consider reflecting the artist's style and touch. Accordingly, we represent reflecting style and touch using morphing technique trained by his own database and introduced especially to generate a rotational in-betweening faces. This database contains center of gravity of each part (right eye, left eye, nose, mouth, eyebrow) and the contours on the facial image. © ACM 2010.

We propose a skinning technique to improve expressive power of Skeleton Subspace Deformation (SSD) by adding the influence of the shape of skeletons to the deformation result by postprocessing. © ACM 2010.

Humans are able to robustly maintain balance in the presence of disturbances by combining a variety of control strategies using posture adjustments and limb motions. Such responses can be applied to balance control in two-armed bipedal robots. We present an upper-body control strategy for improving balance in a humanoid robot. Our method improves on lowerbody balance techniques by introducing an arm rotation strategy (ARS). The ARS uses Q-learning to map sensed state to the appropriate arm control torques. We demonstrate successful balance in a physically-simulated humanoid robot, in response to perturbations that overwhelm lower-body balance strategies alone. © 2010 IEEE.

In this paper, we propose all instant shadow generation technique for 2D animation, especially Japanese Anime. In traditional 2D Anime production, the entire animation including shadows is drawn by hand so that it takes long time to complete. Shadows play all important role in the creation of symbolic visual effects. However shadows are not always drawn due to time constraints and lack of animators especially When the production schedule is tight. To solve this problem, We develop all easy shadowing approach that enables animators to easily create a layer of shadow and its animation based on the character's shapes. Our approach is both instant and intuitive. The only inputs required tire character or object shapes in input animation sequence With alpha value generally used in the Anime production pipeline. First, shadows are automatically rendered on a virtual plane by using a Shadow Map(1) based oil these inputs. Then the rendered shadows call be edited by simple operations and simplified by the Gaussian Filter. Several special effects such as blurring call be applied to the rendered shadow at the same time. Compared to existing approaches, ours is more efficient and effective to handle automatic shadowing in real-time. Copyright (C) 2009 John Wiley & Sons, Ltd.

Skeleton-Subspace Deformation (SSD), which is the most popular method for articulated character animation, often causes some artifacts. Animators have to edit mesh each time, which is seriously tedious and time-consuming. So example based skinning has been proposed. It employs edited mesh as target poses and generates plausible animation efficiently. In this technique, character mesh should be deformed to accurately fit target poses. Mohr et al. [2003] introduced additional joints. They expect animators to embed skeleton precisely.

This paper presents a new methodology for constructing a skin deformation model using MRI and generating accurate skin deformations based on the model. Many methods to generate skin deformations have been proposed and they are classified into three main types. The first type is anatomically based modeling. Anatomically accurate deformations can be reconstructed but computation time is long and controlling generated motion is difficult. In addition, modeling whole body is very difficult. The second is skeleton-subspace deformation (SSD). SSD is easy to implement and fast to compute so it is the most common technique today. However, accurate skin deformations can't be easily realized with SSD. The last type consists of data-driven approaches including example-based methods. In order to construct our model from MRI images, we employ an example-based method. Using examples obtained from medical images, skin deformations can be modeled related to skeleton motions. Retargeting generated motions to other characters is generally difficult with this kind of methods. Kurihara and Miyata realize accurate skin deformations from CT images [Kurihara et al. 2004], but it doesn't mention the possibility of retargeting. With our model, however, generated deformations can be retargeted. Once the model is constructed, accurate skin deformations are easily generated applying our model to a skin mesh. In our experiment, we construct a skin deformation model which reconstructs pronosupination, rotational movement of forearm, and we use range scan data as a skin mesh to apply our model and generate accurate skin deformations.

Characteristics of human motion, such as walking, running or jumping vary from person to person. Differences in human motion enable people to identify oneself or a friend. However, it is challenging to generate animation where individual characters exhibit characteristic motion using computer graphics. Our goal is to construct a system that synthesizes characteristic gait animation automatically. As a result, when crowd animation is generated for instance, the motion with the variation can be made using our system. In our system, we first acquire a silhouette image as input data using a video camera. Second, we extract gait feature from single view silhouette. Finally we automatically synthesize 3D gait animation using the method blending a small number of motion data [KOVAR, L et al 2003].This blending weight is estimated using the gait feature automatically.

The need to rapidly create 3D human head models is still an important issue in game and film production. Blanz et al have developed a morphable model which can semi-automatically reconstruct the facial appearance (3D shape and texture) and simulated hairstyles of "new" faces (faces not yet scanned into an existing database) using photographs taken from the front or other angles [Blanz et al. 2004]. However, this method still requires manual marker specification and approximately 4 minutes of computational time. Moreover, the facial reconstruction produced by this system is not accurate unless a database containing a large variety of facial models is available. We have developed a system that can rapidly generate human head models using only frontal facial range scan data. Where it is impossible to measure the 3D geometry accurately (as with hair regions) the missing data is complemented using the 3D geometry of the template mesh (TM). Our main contribution is to achieve the fast mesh completion for the head modeling based on the "Automatic Marker Setting" and the "Optimized Local Affine Transform (OLAT)". The proposed system generates a head model in approximately 8 seconds. Therefore, if users utilize a range scanner which can quickly produce range data, it is possible to generate a complete 3D head model in one minute using our system on a PC.

In Instant Casting movie System, a personal CG character is automatically generated. The character resembles a participant in a face geometry and texture. However, the voice of a character was an alternative voice determined by the gender of the participant. Therefore sometimes it's not enough to identify the personality of a CG character.In this paper, an automatic pre-scored voice assignment tool for a personal CG character is presented. Voice is essential to identify a personal character as well as a face feature. Our proposed system selects the most similar voice to the participants from voice database, and assigns it as a voice of CG character. Voice similarity criterion is presented by combination of eight acoustic features. After assigning voice data to a personal character, the voice track is played back in synchronization with the movement of the CG character. 60 voice variations have been prepared to our voice database. Validity of the assigned voice has been evaluated by MOS value. The proposed method has achieved 68% of the theoretical figure that is calculated by preliminary experiments.

"Dive into the Movie (DIM)" is a name of project to aim to realize a world innovative entertainment system which can provide an immersion experience into the story by giving a chance to audience to share an impression with his family or friends by watching a movie in which all audience can participate in the story as movie casts. To realize this system, several techniques to model and capture the personal characteristics instantly in face, body, gesture, hair and voice by combining computer graphics, computer vision and speech signal processing technique. Anyway, all of the modeling, casting, character synthesis, rendering and compositing processes have to be performed on real-time without any operator, In this paper, first a novel entertainment system, Future Cast System (FCS), is introduced which can create DIM movie with audience's participation by replacing the original roles' face in a pre-created CG movie with audiences' own highly realistic 3D CG faces. Then the effects of DIM movie on audience experience are evaluated subjectively. The result suggests that most of the participants are seeking for higher realism, impression and satisfaction by replacing not only face part but also body, hair and voice. The first experimental trial demonstration of FCS was performed at the Mitsui-Toshiba pavilion of the 2005 World: Exposition in Aichi Japan. Then, 1,640,000 people have experienced this event during 6 months of exhibition and FCS became one of the most popular events at Expo.2005.

We have developed a visual entertainment system called "Future Cast" which enables anyone to easily participate in a pre-recorded or pre-created film as an instant CG movie star. This system provides audiences with the amazing opportunity to join the cast of a movie in real-time. The Future Cast System can automatically perform all the processes required to make this possible, from capturing participants' facial characteristics to rendering them into the movie. Our system can also be applied to any movie created using the same production process. We conducted our first experimental trial demonstration of the Future Cast System at the Mitsui-Toshiba pavilion at the 2005 World Exposition in Aichi Japan.

We have developed a visual entertainment system called "Future Cast" which enables anyone to easily participate in a pre-recorded or pre-created film as an instant CG movie star. This system provides audiences with the amazing opportunity to join the cast of a movie in real-time. The Future Cast System can automatically perform all the processes required to make this possible, from capturing participants&apos; facial characteristics to rendering them into the movie. Our system can also be applied to any movie created using the same production process. We conducted our first experimental trial demonstration of the Future Cast System at the Mitsui-Toshiba pavilion at the 2005 World Exposition in Aichi Japan.

In this paper we first present all audience-participating movie experience DIM, in which the photo-realistic 3D virtual actor of audience is constructed by computer graphic technologics, and then evaluate the effects of DIM on audience experience using physiological and subjective methods. The empirical results suggest that the participation of virtual actors causes increased subjective sense of presence and engagement, and more intensive emotional responses as compared to traditional movie form: interestingly, there also significantly different physiological responses caused by the participation of virtual actors, objectively indicating the improvement of interaction between audience and movie.

This paper proposes a universal user-friendly post-recording tool for an Instant Casting Movie System (ICS) that enables anyone to be a movie star using his or her own voice and faces. A personal CG character is automatically generated by scanning one's face geometry and image in ICS. Voice is as essential to identify a person as face. However, a character's voice is only based on gender in ICS. We proposed a novel voice recording tool for participants of all ages in a short time. Post-recording tasks are very difficult because speakers should speak in synchronization with the mouth movements of the CG characters. Therefore this task is generally recorded by professional voice actors. Our proposed tool has the following four features: 1) various supporting information for synchronization with voice and mouth movement timing for users
2) automatic post-processing of recorded voices for compositing mixed audio
3) intuitively displays operation for people of all ages
and 4) handles multiple users in parallel for quick recording. We developed a prototype speech synchronization system using a post-recording tool and conducted subjective evaluation experiments of it. Over 60% of the subjects responded that the tool's interface can be operated easily. © 2009 IEEE.

Future cast system is a new entertainment system where participant's face is captured and rendered into the movie as an instant Computer Graphics (CG) movie star, which had been first exhibited at the 2005 World Exposition in Aichi Japan. We are working to add new functionality which enables mapping not only faces but also speech individualities to the cast. Our approach is to find a speaker with the closest speech individuality and apply voice conversion. This paper inves tigates acoustic features to estimate perceptual similarity of speech individuality. We propose a method linearly combined eight acoustic features related to the perception of speech individualities. The proposed method optimizes weights for the acoustic features considering perceptual similarities. We have evaluated performance of our method with Spearman's rank correlation coefficients to perceptual similarities. As the results, the experiments evidenced that the proposed method achieves a correlation coefficient of 0.66.

Creating natural looking hair motion is considered to be one of the most difficult and time consuming challenges in CG animation. A detailed physics-based model is essential in creating convincing hair animation. However, hair animation created using detailed hair dynamics might not always be the result desired by creators. For this reason, a hair simulation system that is both detailed and editable is required in contemporary Computer Graphics. In this paper we therefore, propose the use of External Force Field (EFF) to construct hair motion using a motion capture system. Furthermore, we have developed a system for editing the hair motion obtained using this process. First, the environment around a subject is captured using a motion capture system and the EFF is defined. Second, we apply our EFF-based hair motion editing system to produce creator-oriented hair animation. Consequently, our editing system enables creators to develop desired hair animation intuitively without physical discontinuity.

In this paper we introduce an audience-driven theater experience, DIM Movie, in which audience participates in a pre-created CG movie as its roles, and report the subjective and physiological evaluations for the audience experience offered by DIM movie. Specifically, we present three different experiences to an audience-a traditional movie, its SeIf-DIM (SDIM) version with the audience's participation, and its Self-Friend-DIM (SFDIM) version with co-participation of the audience and his friends. The evaluation results show that the DIM movies (SDIM and SFDIM) elicit greater subjective sense of presence, engagement, and emotional reaction, and stronger physiological response (galvanic skin response, GSR) as compared with the traditional movie form
moreover, audiences show a phasic GSR increase responding to the appearance of their own or friends' CG characters on the movie screen.

In this paper we subjectively and physiologically investigate the effects of the audiences' 3D virtual actor in a movie on their movie experience, using the audience-participating movie DIM as the object of study. In DIM, the photo-realistic 3D virtual actors of audience are constructed by combining current computer graphics (CG) technologies and can act different roles in a pre-rendered CG movie. To facilitate the investigation, we presented three versions of a CG movie to an audience-a Traditional version, its Self-DIM (SDIM) version with the participation of the audience's virtual actor, and its Self-Friend-DIM (SFDIM) version with the coparticipation of the audience and his friends' virtual actors. The results show that the participation of audience's 3D virtual actors indeed cause increased subjective sense of presence and engagement, and emotional reaction
moreover, SFDIM performs significantly better than SDIM, due to increased social presence. Interestingly, when watching the three movie versions, subjects experienced not only significantly different galvanic skin response (GSR) changes on average-changing trend over time, and number of fluctuations-but they also experienced phasic GSR increase when watching their own and friends' virtual 3D actors appearing on the movie screen. These results suggest that the participation of the 3D virtual actors in a movie can improve interaction and communication between audience and the movie. Copyright 2008 ACM.

The role of shadow is important in cartoon animation. Shadows in hand-drawn animation reflect the expression of the animators' style, rather than mere physical phenomena. However shadows in 3DCG cannot express such an animators' touch. In this paper we propose a novel method for editing the shadow with both advantages of hand drawn animation and 3DCG technology. In particular, we discuss two phases that enable animators to transform and deform the shadow tweakably. The first phase is that a shadow projection matrix is applied to deform the shape of the shadow. The second one is that we manipulate vectors to transform the shadow such as scaling and translation. These vectors are used in shadow volume method. The shadows are edited directably by integration of these two phases. Our approach enables animators to edit the shadow by simple mouse operations. As a result, the animators can not only produce shadows automatically but also reflect their style easily. Once the shape and location of shadow are decided by animators' style in our method, 3DCG techniques can produce consistent shadow in object motion interactively.

This paper describes an examination of acoustic features for the estimation of perceptional similarity between speeches. We firstly extract some acoustic features including personality from speeches of 36 persons. Secondly, we calculate each distance between extracted features using Gaussian Mixture Model (GMM) or Dynamic Time Warping (DTW), and then we sort speeches based on the physical similarity. On the other hand, there is the permutation based on the perceptional similarity which is sorted according to the subject. We evaluate the physical features by the Spearman's rank correlation coefficient with two permutations. Consequently, the results show that DTW distance with high STRAIGHT Cepstrum is an optimum feature for estimation of perceptional similarity.

This paper describes a new approach to create cartoon hair animation that allows users to use existing cel character animation sequences. We demonstrate the generation of cartoon hair animation accentuated in 'anime-like' motions. The novelty of this method is that users can choose the existing cel animation of a character's hair animation and apply environmental elements such as wind to other characters With a three-dimensional structure. In fact, users can reuse existing cartoon sequences as input to endow another character with environmental elements as if both characters exist in the same scene. A three-dimensional character's hair motions are created based on hair motions from input cartoon animation sequences. First, users extract hair shapes at each frame from input sequences from Which they then construct physical equations. 'Anime-like' hair motion is created by using these physical equations. Copyright (c) 2006 John Wiley & Soils, Ltd.

An accurate audio-visual speech corpus is inevitable for talking-heads research. This paper presents our audio-visual speech corpus collection and proposes a head-movement normalization method and a facial motion generation method. T he audio-visual corpus contains speech data, movie data on faces, and positions and movements of facial organs. The corpus consists of Japanese phoneme-balanced sentences uttered by a female native speaker. An accurate facial capture is realized by using an optical motion-capture system. We captured high-resolution 3D data by arranging many markers on the speaker's face. In addition, we propose a method of acquiring the facial movements and removing head movements by using affine transformation for computing displacements of pure facial organs. Finally, in order to easily create facial animation from this motion data, we propose a technique assigning the captured data to the facial polygon model. Evaluation results demonstrate the effectiveness of the proposed facial motion generation method and show the relationship between the number of markers and errors.

Head movements could be automatically generated from speech data. The expression of head movement could be also controlled by user-defined emotional factors, as shown in the video demonstration.

Facial animation using the proposed talking heads was created from input speech signals, as shown in the video demonstration. We have confirmed facial animations of various facial objects.

We introduce a multimodal English-to-Japanese and Japanese-to-English translation system that also translates the speaker's speech motion by synchronizing it to the translated speech. This system also introduces both a face synthesis technique that can generate any viseme lip shape and a face tracking technique that can estimate the original position and rotation of a speaker's face in an image sequence. To retain the speaker's facial expression, we substitute only the speech organ's image with the synthesized one, which is made by a 3D wire-frame model that is adaptable to any speaker. Our approach provides translated image synthesis with an extremely small database. The tracking motion of the face from a video image is performed by template matching. In this system, the translation and rotation of the face are detected by using a 3D personal face model whose texture is captured from a video frame. We also propose a method to customize the personal face model by using our GUI tool. By combining these techniques and the translated voice synthesis technique, an automatic multimodal translation can be achieved that is suitable for video mail or automatic dubbing systems into other languages.

This poster proposes a basic idea to observe differences that exist between skeletal postures coming from two methods: postures generated by an only ordinary mocap process, and anatomically and individually accurate skeletal postures generated by a ordinary mocap process together with a medical imaging method such as MRI.

Commercially available motion capture products give us fairly precise movements of human body segments but do not measure enough information to define skeletal posture in its entirety. This sketch describes how to obtain the complete posture of skeletal structure with the help of marker locations relative to bones that are derived from MRI data sets.

A stand-in is a common technique for movies and TV programs in foreign languages. The current stand-in that only substitutes the voice channel results awkward matching to the mouth motion. Videophone with automatic voice translation are expected to be widely used in the near future, which may face the same problem without lip-synchronized speaking face image translation. In this paper, we propose a method to track motion of the face from the video image and then replace the face part or only mouth part with synthesized one which is synchronized with synthetic voice or spoken voice. This is one of the key technologies not only for speaking image translation and communication system, but also for an interactive entertainment system. Finally, an interactive movie system is introduced as an application of entertainment system. © 2003 by Springer Science+Business Media New York.

In recent years, tremendous advances have been achieved in the 3D computer graphics used in the entertainment industry, and in the semiconductor technologies used to fabricate graphics chips and CPUs. However, although good reproduction of facial expressions is possible through 3D CG, the creation of realistic expressions and mouth motion is not a simple task.

HyperMask is a system which projects an animated face onto a physical mask worn by an actor. As the mask moves within a prescribed area, its position and orientation are detected by a camera and the projected image changes with respect to the viewpoint of the audience. The lips of the projected face are automatically synthesized in real time with the voice of the actor, who also controls the facial expressions. As a theatrical tool, HyperMask enables a new style of storytelling. As a prototype system, we put a self-contained HyperMask system in a trolley (disguised as a linen cart), so that it projects onto the mask worn by the actor pushing the trolley.

Synthesizing facial images by computer graphics (CG) has attracted attention in connection with the current trends toward synthesizing virtual faces and realizing communication systems in cyberspace. In this paper, a method for representing human hair, which is known to be difficult to synthesize in computer graphics, is presented. In spite of the fact that hair is visually important in human facial imaging, it has frequently been replaced by simple curved surfaces or a part of the background. Although the methods of representing hair by mapping techniques have achieved results, such methods are inappropriate in representing motions of hair. Thus, spatial curves are used to represent hair, without using textures or polygons. In addition, hair style design is simplified by modeling hair in units of tufts, which are partially concentrated areas of hair. This paper describes the collision decisions and motion representations in this new hair style design system, the modeling of tufts, the rendering method, and the four-branch (quadtree) method. In addition, hair design using this hair style design system and the animation of wind-blown hair are illustrated. © 2002 Scripta Technica, Syst. Comp. Jpn.

Speech-to-speech translation has been studied to realize natural human communication beyond language barriers. Toward further multi-modal natural communication, visual information such as face and lip movements will be necessary In this paper, we introduce a multi-modal English-to-Japanese and Japanese-to-English translation system that also translates the speaker's speech motion while synchronizing it to the translated speech. To retain the speaker's facial expression, we substitute only the speech organ's image with the synthesized one, which is made by a three-dimensional wire-frame model that is adaptable to any speaker. Our approach enables image synthesis and translation with an extremely small database. We conduct subjective evaluation tests using the connected digit discrimination test using data with and without audio-visual lip-synchronization. The results confirm the significant quality of the proposed audio-visual translation system and the importance of lip-synchronization.

Speech-to-speech translation has been studied to realize natural human communication beyond language barriers. Toward further multi-modal natural communication, visual information such as face and lip movements will be necessary. In this paper, we introduce a multi-modal English-to-Japanese and Japanese-to-English translation system that also translates the speaker's speech motion while synchronizing it to the translated speech. To retain the speaker's facial expression, we substitute only the speech organ's image with the synthesized one, which is made by a three-dimensional wire-frame model that is adaptable to any speaker. Our approach enables image synthesis and translation with an extremely small database. We conduct subjective evaluation by connected digit discrimination using data with and without audiovisual lip-synchronicity. The results confirm the sufficient quality of the proposed audio-visual translation system.

We describe the design and report the development of an open source ware toolkit for building an easily customizable anthropomorphic dialog agent. This toolkit consist of four modules for multi-modal dialog integration, speech recognition, speech synthesis, and face image synthesis. In this paper, we focus on the construction of an agent's face image synthesis.

In this paper, a front end system which uses audio and video information to track the people or other sound sources in the ordinary room has developed. The microphone array is used for determining the spatial location of the sound; the active video camera acquires the image of the area where the sound is detected, detects the people in the image by using skin color and can zoom and track a speaker. Several add-ons to the system include various visualization tools such as on-screen displays of waveforms, correlation plots, spectrum plots, spatial acoustic energy distribution, running time-frequency acoustic energy plots, and the possibility of real-time beamforming with real-time output to the headphones. The system can be used as a front-end for the non-encumbering human-computer interaction by video and audio means.

In this paper, we introduce a multi-modal English-to-Japanese and Japanese-to-English translation system that also translates the speaker's speech motion while synchronizing it to the translated speech. To retain the speaker's facial expression, we substitute only the speech organ's image with the synthesized one, which is made by a three-dimensional wire-frame model that is adaptable to any speaker. Our approach enables image synthesis and translation with an extremely small database.

A stand-in is a common technique for movies and TV programs in foreign languages. The current stand-in that only substitutes the voice channel results awkward matching to the mouth motion. Videophone with automatic voice translation are expected to be widely used in the near future, which may face the same problem without lip-synchronized speaking face image translation. In this paper, we propose a method to track motion of the face from the video image, that is one of the key technologies for speaking image translation. Almost all the old tracking algorithms aim to detect feature points of the face. However, these algorithms had problems, such as blurring of a feature point between frames and occlusion of the hidden feature point by rotation of the head, and so on. In this paper, we propose a method which detects movement and rotation of the head given the three dimensional shape of the face, by template matching using a 3D personal face wireframe model. The evaluation experiments are carried out with the measured reference data of the head. The proposed method achieves 0.48 angle error in average. This result confirms effectiveness of the proposed method.

HYPERMASK is a system which projects an animated face onto a physical mask, worn by an actor. As the mask moves within a prescribed area, its position and orientation are detected by a camera, and the projected image changes with respect to the viewpoint of the audience. The lips of the projected face are automatically synthesized in real time with the voice of the actor, who also controls the facial expressions. As a theatrical tool, HYPERMASK enables a new style of storytelling. As a prototype system, we propose to put a self-contained HYPERMASK system in a trolley (disguised as a linen cart), so that it projects onto the mask worn by the actor pushing the trolley.

Muscle based face image synthesis is one of the most realistic approaches to the realization of a life-like agent in computers. A facial muscle model is composed of facial tissue elements and simulated muscles. In this model, forces are calculated effecting a facial tissue element by contraction of each muscle string, so the combination of each muscle contracting force decides a specific facial expression. This muscle parameter is determined on a trial and error basis by comparing the sample photograph and a generated image using our Muscle-Editor to generate a specific face image. In this paper, we propose the strategy of automatic estimation of facial muscle parameters from 2D markers' movements located on a face using a neural network. This corresponds to the non-realtime 3D facial motion capturing from 2D camera image under the physics based condition.

This paper proposes a new real-time method for estimating human postures in 3D from trinocular images. In this method, an upper body orientation detection and a heuristic contour analysis are performed on the human silhouettes extracted from the trinocular images so that representative points such as the top of the head can be located. The major joint positions are estimated based on a genetic algorithm-based learning procedure. 3D coordinates of the representative points and joints are then obtained from the two views by evaluating the appropriateness of the three views. The proposed method implemented on a personal computer runs in real-time. Experimental results show high estimation accuracies and the effectiveness of the view selection process. © 2000 IEEE.

Muscle based face image synthesis is one of the most realistic approach to realize life-like agent in computer. Facial muscle model is composed of facial tissue elements and muscles. In this model, forces are calculated effecting facial tissue element by contraction of each muscle strength, so the combination of each muscle parameter decide a specific facial expression. Now each muscle parameter is decided on trial and error procedure comparing the sample photograph and generated image using our Muscle-Editor to generate a specific face image.
In this paper we propose the strategy of automatic estimation of facial muscle parameters from 2D marker movements using neural network.
This is also 3D motion estimation from 2D point or flow information in captured image under restriction of physics based face model.

This paper introduces a new real-lime method to estimate the posture of a human from thermal images acquired by an infrared camera regardless of the background and lighting conditions. Distance transformation is performed for the human body area extracted from the thresholded thermal image for the calculation of the center of gravity. After the orientation of the zipper half of the body is obtained by calculating the moment of inertia, significant points such as the top of the head, the tips oft he hands and foot are heuristically located. In addition, the elbow and knee positions are estimated from the detected (significant) points using a genetic algorithm based learning procedure.
The experimental results demonstrate the robustness of the proposed algorithm and real-time (faster than 20 frames per second) performance.

Muscle based face image synthesis is one of the most realistic approach to realize life-like agent in computer.
Facial muscle model is composed of facial tissue elements and muscles. In this model, forces are calculated effecting facial tisssue element by contraction of each muscle strength, so the combination of each muscle parameter decide a specific facial expression.
Now each muscle parameter is decided on trial and error procedure comparing the sample photograph and generated image using our Muscle-Editor to generate a specific face image. In this paper, we propose the strategy of automatic estimation of facial muscle parameters front 2D marker movements using neural network. This corresponds to the non-realtime 3D facial motion tracking from 2D image under the physics based condition.

Muscle based face image synthesis [1] [2] is one of the most realistic approach to realize interface agent. Facial muscle model is composed of facial tissue elements and muscles. In this model, forces affecting facial tisssue element are calculated by contraction of each muscle strength, so the combination of each muscle parameter decide a specific facial expression.
In this paper, we introduce the facial muscle model and the strategy of automatic estimation of facial muscle parameters from 2-D marker movements.

This paper introduces a new;real-time method to estimate the posture of a human from thermal images acquired by an infrared camera regardless of the background and lighting conditions. Distance transformation is performed for the human body area extracted from the thresholded thermal image for the calculation of the center of gravity. After the orientation of the upper half of the body is obtained by calculating the moment of inertia, significant points such as the top of the head, the tips of the hands and foot are heuristically located. In addition, the elbow and knee positions are estimated from the detected (significant) points using a genetic algorithm based learning procedure.
The experimental results demonstrate the robustness of the proposed algorithm and rear-time (faster than 20 frames per second) performance.

A new facial feature extraction technique for expression recognition is proposed. We employ the spatial frequency domain information to obtain robust performance to the random noise on a image or the lighting conditions. It exhibited high ability sufficiently even if combined with a low-performance region tracking method. As an application of this technique, we have constructed a dynamic facial expression recognition system. We use hidden Markov models to utilize temporal changes in the facial expressions. The spatial frequency information and the temporal information make better rates of facial expression recognition. In the experiment, we established a correct response rate of approximately 84.1% of recognition with six categories. © 1996 IEEE.

The goal of research is to realize natural human-machine communication system by givig a facial expression to computer. To realize this environment, it′s necessary for the machine to recognize human′s emotion condition appearing in the human′ face, and synthesize the reasonable facial image of machine. For this purpose, the machine should have emotion model based on parameterized faces which can express an emotion conditon quantitatively. Especially, both mapping and inverse mapping from face to the model have to be achieve. Facial expression of parameterized with Facial Action Coding System (FACS) which is translated to the grids′ motion of face model. An emotion condition is described compactly by the pint in a 3D space generated by 5-layered neural network and its evaluation result shows the high performance of this model.

In this paper, the methods for modeling racial expression and emotion are proposed. This Emotion Model, called 3-D Emotion Space can represent both human and computer emotion conditions appearing on the face as a coordinate in the 3-D Space.
For the construction of this 3-D Emotion Space, 5-layer neural network which is superior in non-linear mapping performance is applied. After the network training with backpropagation to realize Identity Mapping, both mapping from facial expression parameters to the 3-D Emotion Space and inverse mapping from the Emotion Space to the expression parameters were realized.
As a result a system which can analyze acid synthesize the facial expression were constructed simultaneously.
Moreover, this inverse mapping to the facial expression is evaluated by the subjective evaluation using the synthesized expressions as Lest images. This evaluation result proved the high performance to describe natural facial expression and emotion condition with this model.

A trial to generate the object in the natural world by computer graphics is now actively done. Normally, huge computing power and storage capacity are necessary to make real and natural movement of the human's hair. In this paper, a technique to synthesize human's hair with short processing time and little storage capacity is discussed. A new Space Curve Model and Rigid Segment Model are proposed. And also high speed collision detection with the human's body is discussed.

We're building "face" interface which is a user-friendly human-machine interface with Multi-media and can realize face-to-face communication environment between an operator and a machine. In this system, human natural face appears on the display of machine and can talk to operator with natural voice. This paper describes face motion and expression synthesis schemes which can be applied to this "face" interface. We express a human head with 3D model. The surface model is built by texture mapping with 2D real image. All the motions and expressions are synthesized and controlled automatically by the movement of some feature points on the model. This "face" interface is one of the applications of model based image coding and media conversion schemes we've been studying.

An automatic facial motion image synthesis scheme, driven by speech, and a real-time image synthesis design are presented. The purpose of this research is to realize an "intelligent" human-machine interface or "intelligent" communication system with talking head images. A human face is reconstructed on the display of a terminal using a 3-D surface model and texture mapping technique. Facial motion images are synthesized naturally by transformation of the lattice points on 3-D wire frames. Two driving motion methods, a text-to-image conversion scheme, and a voice-to-image conversion scheme are proposed in this paper. In the first method, the synthesized head image can appear to speak some given words and phrases naturally. In the latter case, some mouth and jaw motions can be synthesized in synchronization with voice signals from a speaker. Facial expressions, other than mouth shape and jaw position, also can be added at any moment, so it is easy to make the facial model appear angry, to smile, to appear sad, etc., by special modification rules. These schemes were implemented on a parallel image computer system. A real-time image synthesizer was able to generate facial motion images on the display, at a TV image video rate.

Model‐based image coding has been highlighted recently as a high‐efficiency coding method for TV telephone and TV conference systems. In a model‐based coding system, an ultralow‐rate image transmission is realized by obtaining common models of a facial image at both sides of a communication and by transmitting only modification parameters between them. However, it is difficult to realize a realtime processing of a model‐based coding with a conventional iterative‐processing type computer since the amount of material to analyze and synthesize at both sides is very large. Realtime processing is absolutely necessary to realize a practical system using this method, i.e., highspeed processings at both the transmitter and receiver are required. This paper describes a realtime facial image synthesis method based on a multiprocessor construction with a transputer. The transputer is a microcomputer having a communication capability with multiple processors and 10 MIPS CPU performance. Using this system in a pipeline processing with 20 processors, it is possible to synthesize realtime facial images at a rate of about 16 frames per second. If only a part around the lips is transmitted, it is possible to synthesize an image at a rate of about 40 frames per second with a network using five processors. Copyright © 1991 Wiley Periodicals, Inc., A Wiley Company

A facial motion image synthesis method for intelligent man‐machine interface is examined. Here, the intelligent man‐machine interface is a kind of friendly man‐machine interface with voices and pictures in which human faces appear on a screen and answer questions, compared to the currently existing user interfaces which primarily uses letters. Thus what appears on the screen is human faces, and if speech mannerisms and facial expressions are natural, then the interactions with the machine are similar to those with actual human beings. To implement such an intelligent man‐machine interface it is necessary to synthesize natural facial expressions on the screen. This paper investigates a method to synthesize facial motion images based on given information on text and emotion. The proposed method utilizes the analysis‐synthesis image coding method. It constructs facial images by assigning intensity data to the parameters of a 3‐dimensional (3‐D) model matching the person in question. Moreover, it synthesizes facial expressions by modifying the 3‐D model according to the predetermined set of rules based on the input phonemes and emotion, and also synthesizes reasonably natural facial images. Copyright © 1991 Wiley Periodicals, Inc., A Wiley Company

Generally speaking, it is necessary to extract knowledge from an expert in a given discipline and implement this knowledge into a system when constructing an expert system. However, it is not easy to extract knowledge in such fields as medical diagnosis or pattern recognition because the inference logic depends on the experience and intuition of the expert. This paper proposes an automatic rule extraction mechanism using statistical analysis. In this system, production rules are expressed in the form of the threshold function. Because the threshold function can describe any kinds of inference logic, it is expressed easily as a linear combination of input vectors and weighting coefficients. Thus weighting coefficients can be calculated by the same method as a discriminant function. If only one threshold is defined, general Boolean logic an be expressed. Moreover, an ambiguous inference rule can be expressed when the threshold levels are multidefined and a membership function is defined at each category. Further, the Fuzzy Tree Search algorithm which combines ambiguous inference and tree search is proposed at the end of this paper. This algorithm can search and determine an optimum cluster with little calculation and good performance. In practice, a medical diagnostic system applied to psychiatry which has most ambiguous diagnostic logic, has been constructed based on this algorithm and inference rules have been extracted automatically. By this experiment, Fuzzy Tree Search is as fast as the tree search technique and has as good a performance as a full search clustering technique. Copyright © 1988 Wiley Periodicals, Inc., A Wiley Company