Nuclei segmentation is fundamental and crucial for analyzing histopathological images. Generally, a pathological image contains tens of thousands of nuclei, and there exists clustered nuclei, so it is difficult to separate each nucleus accurately. Challenges against blur boundaries, inconsistent staining, and overlapping regions have adverse effects on segmentation performance. Besides, nuclei from various organs appear quite different in shape and size, which may lead to the problems of over-segmentation and under-segmentation. In order to capture each nucleus on different organs precisely, characteristics about both nuclei and boundaries are of equal importance. Thus, in this article, we propose a contextual mixing feature Unet (CMF-Unet), which utilizes two parallel branches, nuclei segmentation branch and boundary extraction branch, and mixes complementary feature maps from two branches to obtain rich and integrated contextual features. To ensure good segmentation performance, a multiscale kernel weighted module (MKWM) and a dense mixing feature module (DMFM) are designed. MKWM, used in both nuclei segmentation branch and boundary extraction branch, contains a multiscale kernel block to fully exploit characteristics of images and a weight block to assign more weights on important areas, so that the network can extract discriminative information efficiently. To fuse more beneficial information and get integrated feature maps, the DMFM mixes the feature maps produced by the MKWM from two branches to gather both nuclei information and boundary information and links the feature maps in a densely connected way. Because the feature maps produced by the MKWM and DMFM are both sent into the decoder part, segmentation performance can be enhanced effectively. We test the proposed method on the multi-organ nuclei segmentation (MoNuSeg) dataset. Experiments show that the proposed method not only performs well on nuclei segmentation but also has good generalization ability on different organs.

Abstract

Various deep neural network architectures (DNNs) maintain massive vital records in computer vision. While drawing attention worldwide, the design of the overall structure lacks general guidance. Based on the relationship between DNN design and numerical differential equations, we performed a fair comparison of the residual design with higher order perspectives. We show that the widely used DNN design strategy, constantly stacking a small design (usually, 2–3 layers), could be easily improved, supported by solid theoretical knowledge and with no extra parameters needed. We reorganise the residual design in higher order ways, which is inspired by the observation that many effective networks can be interpreted as different numerical discretisations of differential equations. The design of ResNet follows a relatively simple scheme, which is Euler forward; however, the situation becomes complicated rapidly while stacking. We suppose that stacked ResNet is somehow equalled to a higher order scheme; then, the current method of forwarding propagation might be relatively weak compared with a typical high-order method such as Runge–Kutta. We propose HO-ResNet to verify the hypothesis on widely used CV benchmarks with sufficient experiments. Stable and noticeable increases in performance are observed, and convergence and robustness are also improved. Our stacking strategy improved ResNet-30 by 2.15% and ResNet-58 by 2.35% on CIFAR-10, with the same settings and parameters. The proposed strategy is fundamental and theoretical and can, therefore, be applied to any network as a general guideline.

Graphical abstract

Modern image classifiers are often suffering over-fitting problems because of the insufficient number of images in the dataset. Data augmentation is a strategy to increase the number of training samples. However, recent data augmentation methods are designed manually and cannot generate real-like images. Some neural network-based image generation methods such as GAN and VAE can also be used for data augmentation, but they are usually applied to unbalanced datasets. Since the generated images cannot be guaranteed to be from the same label, using them to extend a balanced dataset may lead to decreasing the accuracy of the classifier. In this paper, we propose an image transfer network to produce images that automatically adapt to a specific dataset and classifier. The image transfer network will search for the output images which can maximize the validation accuracy and help the classifier to overcome the over-fitting problems. Through the experiments, our method achieves high accuracy on CIFAR-10 and CIFAR-100 datasets. Moreover, since it could combine with other data augmentation methods, we show that using our method can push the state-of-the-art results furthermore.

© 2018 APSIPA organization. This paper presents an accelerated constant-time Gaussian filter (O(1) GF) specialized in short window length where constant-time (O(1)) means that computational complexity per pixel does not depend on filter window length. Our method is extensively designed based on the idea of O(1) GF based on Discrete Cosine Transform (DCT). This framework approximates a Gaussian kernel by a linear sum of cosine terms and then convolves each cosine term in O(1) per pixel using sliding transform. Importantly, if window length is short, DCT-1 consists of easily-computable cosine values such as 0, \pm\frac{1}{2} and ±1. This behavior is not satisfied in other DCT types. From this fact, our method accelerates the sliding transform by employing DCT-1 focusing on short window length. Experiments show that our method overcomes naive Gaussian convolution and existing O(1) GF in terms of computational time. Interestingly, the results also reveal that, without truncating negligible terms, our method runs faster than convolution.

With the rapid development of multimedia technology, it's easy for someone to obtain an image and edit it according to their own preferences or some ulterior purpose. Copy-Move is a common type of digital image forgery where a part of the original image is copied and pasted at another position in the same image. In this paper, we propose an efficient methodology for enhancing block matching based on Copy-Move forgery detection. The main contribution of this work is the utilization of polar representation to get the representative features for each block. The main feature used in this paper is the frequency of each block based on Fourier transform. The experimental results show the efficiency of the proposed method for detecting copy-move regions, even when the copied region has undergone severe image manipulations such as rotation, scaling, Gaussian blurring, brightness modification, JPEG compression and noise addition.

In this paper, we propose an infinite impulse response (IIR) filtering with complex coefficients for Euclid distance based filtering, e.g. bilateral filtering. Recursive filtering of edge-preserving filtering is the most efficient filtering. Recursive bilateral filtering and domain transform filtering belong to this type. These filters measure the difference between pixel intensities by geodesic distance. Also, these filters do not have separability. The aspects make the filter sensitive to noises. Bilateral filtering does not have these issues, but it is time-consuming. In this paper, edge-preserving filtering with the complex exponential function is proposed. The resulting stack of these IIR filtering is merged to approximated edge-preserving in FIR filtering, which includes bilateral filtering. For bilateral filtering, a raised-cosine function is used for efficient approximation. The experimental results show that the proposed filter, named IIR bilateral filter, approximates well and the computational cost is low.

Seed detection or sometimes known as nuclei detection is a prerequisite step of nuclei segmentation which plays a critical role in quantitative cell analysis. The detection result is considered as accurate if each detected seed lies only in one nucleus and is close to the nucleus center. In previous works, voting methods are employed to detect nucleus center by extracting the nucleus saliency features. However, these methods still encounter the risk of false seeding, especially for the heterogeneous intensity images. To overcome the drawbacks of previous works, a novel detection method is proposed, which is called secant normal voting. Secant normal voting achieves good performance with the proposed skipping range. Skipping range avoids over-segmentation by preventing false seeding on the occlusion regions. Nucleus centers are obtained by mean-shift clustering from clouds of voting points. In the experiments, we show that our proposed method outperforms the comparison methods by achieving high detection accuracy without sacrificing the computational efficiency.

This paper proposed a disparity refinement method based on two-stage cross. First stage is anti-texture cross-based support region construction to build proper support regions for error pixels without being influenced by texture. Based on the support regions, second stage of the method is proposed, which is called weighted cross-based updating method. The experiments show that the proposed method could build the support region accurately and improve the accuracy of the disparity map in final results with fast speed, compared to other tree-based algorithms. It also outperforms the existing disparity refinement methods in preserving the boundaries of objects in the final disparity map.

Microscopic images are quite helpful for us to observe the details of cells because of its high resolution. Furthermore it can benefit biologists and doctors to view the cell structure from any aspect by using a serial images to generate 3D cell structure. However each cell slice is placed at the microscopy respectively, which will bring in the arbitrary rotation and translation among the serial slices. What's more, the sectioning process will destroy the cell structure such as tearing or warping. Therefore we must register the serial slices before rendering the volume data in 3D. In this paper we propose a robust registration algorithm based on an improved 3D Hilbert scam search. Besides we put forward a simple but effective method to remove false matching in consecutive images. Finally we correct the local deformation based on optical-flow theory and adopt multi-resolution method. Our algorithm is tested, on a serial microscopy kidney cell images, and the experimental results show how accurate and robust of our method is.

In this paper, a novel method called discriminative histogram intersection metric learning (DHIML) is proposed for pair matching and classification. Specifically, we introduce a discrimination term for learning a metric from binary information such as same/not-same or similar/dissimilar, and then combine it with the classification error for the discrimination in classifier construction. Compared with conventional approaches, the proposed method has several advantages. 1) The histogram intersection strategy is adopted into metric learning to deal with the widely used histogram features effectively. 2) By introducing discriminative term and classification error term into metric learning, a more discriminative distance metric and a classifier can be learned together. 3) The objective function is robust to outliers and noises for both features and labels in the training. The performance of the proposed method is tested on four applications: face verification, face-track identification, face-track clustering, and image classification. Evaluations on the challenging restricted protocol of Labeled Faces in the Wild (LFW) benchmark, a dataset with more than 7 000 face-tracks, and Caltech-101 dataset validate the robustness and discriminability of the proposed metric learning, compared with the recent state-of-the-art approaches.

Recent works have achieved near or over human performance in traditional face recognition under PIE (pose, illumination and expression) variation. However, few works focus on the cross-age face recognition task, which means identifying the faces from same person at different ages. Taking human-aging into consideration broadens the application area of face recognition. It comes at the cost of making existing algorithms hard to maintain effectiveness. This paper presents a new reference based approach to address cross-age problem, called Eigen-Aging Reference Coding (EARC). Different from other existing reference based methods, our reference traces eigen faces instead of specific individuals. The proposed reference has smaller size and contains more useful information. To the best of our knowledge, we achieve state-of-the-art performance and speed on CACD dataset, the largest public face dataset containing significant aging information.

This paper presents an efficient constant-time bilateral filter where constant-time means that computational complexity is independent of filter window size. Many state-of-the-art constant-time methods approximate the original bilateral filter by an appropriate combination of a series of convolutions. It is important for this framework to optimize the performance tradeoff between approximate accuracy and the number of convolutions. The proposed method achieves the optimal performance tradeoff in a least-squares manner by using spectral decomposition under the assumption that images consist of discrete intensities such as 8-bit images. This approach is essentially applicable to arbitrary range kernel. Experiments show that the proposed method outperforms state-of-the-art methods in terms of both computational complexity and approximate accuracy.

In this study, we propose a simple, yet general and powerful framework of integrating multiple global and local features by Product Sparse Coding (PSC) for image retrieval. In our framework, multiple global and local features are extracted from images and then are transformed to Trimmed-Root (TR)-features. After that, the features are encoded into compact codes by PSC. Finally, a two-stage ranking strategy is proposed for indexing in retrieval. We make three major contributions in this study. First, we propose TR representation of multiple image features and show that the TR representation offers better performance than the original features. Second, the integrated features by PSC is very compact and effective with lower complexity than by the standard sparse coding. Finally, the two-stage ranking strategy can balance the efficiency and memory usage in storage. Experiments demonstrate that our compact image representation is superior to the state-of-the-art alternatives for large-scale image retrieval.

This paper presents an efficient constant-time bilateral filter where constant-time means that computational complexity is independent of filter window size. Many state-of-the-art constant-time methods approximate the original bilateral filter by an appropriate combination of a series of convolutions. It is important for this framework to optimize the performance tradeoff between approximate accuracy and the number of convolutions. The proposed method achieves the optimal performance tradeoff in a least-squares manner by using spectral decomposition under the assumption that images consist of discrete intensities such as 8-bit images. This approach is essentially applicable to arbitrary range kernel. Experiments show that the proposed method outperforms state-of-the-art methods in terms of both computational complexity and approximate accuracy.

Color space is one of the bases in the image processing area. Suitable color space can give the suitable description of colors for variant processing. However, in the image processing area, the existing color space cannot show the suitable distribution in color and lightness. In this paper, a novel color space based on RGB color barycenter (RGB-CB) is proposed to describe the color and lightness more intuitively. To prove the effectiveness of the proposed color space, YUV, HSV, L* a* b*, and IPT color spaces are discussed and compared. Experimental results show the proposed color space can perform better effect than other color space in image processing.

This paper addresses the problem of scene classification and proposes learning discriminative and shareable patches (LDSP) method. The main idea of learning discriminative and shareable patches is to discover patches that exhibit both large between-class dissimilarity (discriminative) and large within-class similarity (shareable). A novel and efficient re-clustering, based on co-occurrence relationship of first-step clustering, is proposed and conducted to further enhance the visual similarity of patches within each cluster. In order to establish appropriate criteria for selecting desired patches, a condensed representation of image features called feature epitome is introduced. In the classification, a patch feature involving pre-trained convolutional neural network model is investigated. The experimental result outperforms existing single-feature methods on MIT 67 scene benchmark in term of mean Accuracy Precision.

Keypoint detection and description using approximate continuous scale space are more efficient techniques than typical discretized scale space for achieving more robust feature matching. However, this state-of-the-art method requires high computational complexity to approximately reconstruct, or decompress, the value at an arbitrary point in scale space. Specifically, it has O(M-2) computational complexity where M is an approximation order. This paper presents an efficient scale space approach that provides decompression operation with O(M) complexity without a loss of accuracy. As a result of the fact that the proposed method has much fewer variables to be solved, the least-square solution can be obtained through normal equation. This is easier to solve than the existing method which employs Karhunen-Loeve expansion and generalized eigenvalue problem. Experiments revealed that the proposed method performs as expected from the theoretical analysis.

Magnetic Resonance Imaging (MRI) has been widely used in medical diagnose because of its non-invasive manner and excellent depiction of soft-tissue changes. Recently, the compressive sensing (CS) theory has been applied to reconstruct the MR image from highly down-sampled k-space data, which can reduce the scanning duration. To obtain useful information as much as possible with the same sampling rate, a weighted sampling strategy is studied. Moreover, based on the advantage of CS, a Wavelet tree based reconstruction approach is proposed. The experimental results demonstrate that the proposed method is preferable to other methods.

In this paper, we proposed an optimized Sparse Deep Learning Network (SDLN) model for Face Recognition (FR). A key contribution of this work is to learn feature coding of human face with a SDLN based on local structured Sparse Representation (SR). In traditional sparse FR methods, different poses and expressions of training samples could have great influence on the recognition results. We consider the SR that should be guided by context constraints which are defined by the correlations of dictionary atoms. The over complete common dictionary that contains common atom set has been learned from a local region structured sparse encoding process. We obtained over-complete common dictionary and feature coding for each face. As we all know that the deep learning has been widely applied to face feature learning. Using traditional deep learning methods can not contain variations of face identity information. We have to get face features of compatible change in a jointly deep learning network. The proposed SDLN is jointly fine-tuned to optimize for the task of FR. The SDLN achieves high FR performance on the ORL and FERET database.

Keypoint detection and description in a continuous scale space achieves better robustness to scale change than those in a discretized scale space. State-of-the-art methods first decompose a continuous scale space into M + 1 component images weighted by M-order polynomials of scale sigma, and then reconstruct the value at an arbitrary point in the scale space by a linear combination of the component images. However, these methods create the mismatch that, if sigma is large, common filter kernels such as Gaussian converge to zero; but the polynomials of sigma diverge. This paper presents a more efficient approximation that suppresses this mismatch by normalizing the weighting functions. Experiments show that the proposed method achieves higher performance trade-off than state-of-the-art methods: it reduces the number of component images and total running time by 20-40% without a sacrifice of stability in keypoints detection.

This paper presents an acceleration method of the bilateral filter (BF) for multi-channel images. In most existing acceleration methods, the BF is approximated by an appropriate combination of convolutions. A major purpose under this framework is to achieve sufficient approximate accuracy by as few convolutions as possible. However, state-of-the-art methods for multi-channel images still requires hundreds of (e.g., 256) convolutions to achieve sufficient accuracy. The proposed method reduces the number of convolutions without a loss in accuracy via soft-assignment coding. This approach enables us to take two major advantages that two state-of-the-art methods (scalar quantization with linear interpolation and vector quantization) have individually provided. Experiments show that the proposed method can produce sufficiently-accurate resulting images by using 64-80 convolutions only.

With rapidly advancing of contemporary medicine, it is necessary to help people identify various kinds of pills to prevent the adverse pill events. In this study, a high-accuracy automatic pill recognition system is proposed for accurate and automatic pill recognition. As pill imprint is main distinction between different pills, this system proposes algorithms on both imprint extraction and description parts to make use of imprint information. First, proposed modified stroke width transform is adopted to extract the imprint by detecting coherent strokes of imprint on the pill. Moreover, image segmentation by Loopy belief propagation is also added on printed imprint pills to solve the incoherent and coarse stroke problem. Second, a new descriptor named two-step sampling distance sets is proposed for accurate imprint description and successfully cut down the noise on extracted imprint. This strategy is based on the imprint partition - partitions the imprint on the basis of separated strokes, fragments and noise points. Recognition experiments are applied on extensive databases and result shows 90.46% rank-1 matching accuracy and 97.16% on top five ranks when classifying 12 500 query pill images into 2500 categories.

This paper presents an efficient constant-time bilateral filter that produces a near-optimal performance tradeoff between approximate accuracy and computational complexity without any complicated parameter adjustment, called a compressive bilateral filter (CBLF). The constant-time means that the computational complexity is independent of its filter window size. Although many existing constant-time bilateral filters have been proposed step-by-step to pursue a more efficient performance tradeoff, they have less focused on the optimal tradeoff for their own frameworks. It is important to discuss this question, because it can reveal whether or not a constant-time algorithm still has plenty room for improvements of performance tradeoff. This paper tackles the question from a viewpoint of compressibility and highlights the fact that state-of-the-art algorithms have not yet touched the optimal tradeoff. The CBLF achieves a near-optimal performance tradeoff by two key ideas: 1) an approximate Gaussian range kernel through Fourier analysis and 2) a period length optimization. Experiments demonstrate that the CBLF significantly outperforms state-of-the-art algorithms in terms of approximate accuracy, computational complexity, and usability.

This paper presents an efficient constant-time algorithm for Gaussian filtering and also Gaussian derivative filtering that provides a high approximate accuracy in a low computational complexity regardless of its filter window size. The proposed algorithm consists of two key techniques: second-order shift properties of the Discrete Cosine/Sine Transforms type-5 and dual-domain error minimization for finding optimal parameters. The former enables us to perform filtering in fewer number of arithmetic operations as compared than some state-of-the-art algorithms without integral images. The latter enables us to find the optimal filter size that provides the most accurate filter kernel approximation. Experiments show that the proposed algorithm clearly outperforms state-of-the-art ones in computational complexity, approximate accuracy, and accuracy stability.

© 2015 by ITE Transactions on Media Technology and Applications (MTA). In this paper, a contrast enhancement method called Multi-Histogram Mapping and Fusion (MHMF) is proposed for color images. Histogram analysis based method has been successfully applied to contrast enhancement in some applications, but they are hard to enhance the dark and bright regions simultaneously for back-light images. To solve this problem, the color barycenter model (CBM) is extended to separate the color image into lightness and chroma components. Then multi-histogram mapping (MHM) is used to map the lightness component of one single color image into several new lightness components with different contrast. These new components are divided into several patches and the best patches are selected by calculating image entropy. Finally, the selected patches are fused to create the enhanced image, and mix-Gaussian filter is applied to remove the sharp transition. The experimental results show the effectiveness of proposed method comparing with other state-of-the-art methods.

This paper presents an efficient constant-time algorithm for Gaussian filtering and also Gaussian derivative filtering that provides a high approximate accuracy in a low computational complexity regardless of its filter window size. The proposed algorithm consists of two key techniques: second-order shift properties of the Discrete Cosine/Sine Transforms type-5 and dual-domain error minimization for finding optimal parameters. The former enables us to perform filtering in fewer number of arithmetic operations as compared than some state-of-the-art algorithms without integral images. The latter enables us to find the optimal filter size that provides the most accurate filter kernel approximation. Experiments show that the proposed algorithm clearly outperforms state-of-the-art ones in computational complexity, approximate accuracy, and accuracy stability.

The recent years has seen immense improvement in the development of signal processing based on Curvelet transform. The Curvelet transform provide a new multi-resolution representation. The frame elements of Curvelets exhibit higher direction sensitivity and anisotropic than the Wavelets, multi-Wavelets, steerable pyramids, and so on. These features are based on the anisotropic notion of scaling. In practical instances, time series signals processing problem is often encountered. To solve this problem, the time-frequency analysis based methods are studied. However, the time-frequency analysis cannot always be trusted. Many of the new methods were proposed. The Empirical Mode Decomposition (EMD) is one of them, and widely used. The EMD aims to decompose into their building blocks functions that are the superposition of a reasonably small number of components, well separated in the time-frequency plane. And each component can be viewed as locally approximately harmonic. However, it cannot solve the problem of directionality of high-dimensional. A reallocated method of Curvelet transform (optimized Curvelet-based EMD) is proposed in this paper. We introduce a definition for a class of functions that can be viewed as a superposition of a reasonably small number of approximately harmonic components by optimized Curvelet family. We analyze this algorithm and demonstrate its results on data. The experimental results prove the effectiveness of our method.

This paper proposes a novel method for estimating disparity accurately. To achieve the ideal result, an optimal adjusting framework is proposed to address the noise, occlusions, and outliners. Different from the typical multi-view stereo (MVS) methods, the proposed approach not only use the color constraint, but also use the geometric constraint associating multiple frame from the image sequence. The result shows the disparity with a good visual quality that most of the noise is eliminated, the errors in occlusion area are suppressed and the details of scene objects are preserved.

Dynamic MRI is widely used for many clinical exams but slow data acquisition becomes a serious problem. The application of Compressed Sensing (CS) demonstrated great potential to increase imaging speed. However, the performance of CS is largely depending on the sparsity of image sequence in the transform domain, where there are still a lot to be improved. In this work, the sparsity is exploited by proposed Sparse Decomposition Learning (SDL) algorithm, which is a combination of low-rank plus sparsity and Blind Compressed Sensing (BCS). With this decomposition, only sparsity component is modeled as a sparse linear combination of temporal basis functions. This enables coefficients to be sparser and remain more details of dynamic components comparing learning the whole images. A reconstruction is performed on the undersampled data where joint multicoil data consistency is enforced by combing Parallel Imaging (PI). The experimental results show the proposed methods decrease about 15 similar to 20% of Mean Square Error (MSE) compared to other existing methods.

This paper proposes a disparity refinement method with stability-based tree. By developing stability-based tree to evaluate and reconstruct support regions for error parts, the proposed method achieves effective performance in removing outliers. This approach further improves the quality of raw disparity map in stereo matching, which makes the local methods results comparable to the global ones. Experiments exhibit that the proposed method reduces more than 70% aggregation time compared with traditional tree method without loss of accuracy. It also outperforms existing disparity refinement methods in removing large error parts.

© 2015 The Society of Instrument and Control Engineers-SICE. In the city of Kitakyushu, a quarter more of people are older than 65 years old. Especially a large proportion of elderly people are living on their own. So how to let old people easily access the communal facilities like hospitals need be attended. Therefore, the bad connection between public transportation and home is need to be solved. Based on the survey result and road situation in Kitakyushu city, a new type small single-seat electrical vehicle (sEV) is studied to provide a solution for elder to easily access the public transportation. In this research, low cost and safe automatic driving electrical vehicle based on limited number of sensors is focused.

Road lane detection is a key problem in advanced driver-assistance systems (ADAS). For solving this problem, vision-based detection methods are widely used and are generally focused on edge information. However, only using edge information leads to miss detection and error detection in various road conditions. In this paper, we propose a neighbor-based image conversion method, called extremal-region enhancement. The proposed method enhances the white lines in intensity, hence it is robust to shadows and illuminance changes. Both edge and shape information of white lines are extracted as lane features in the method. In addition, we implement a robust road lane detection algorithm using the extracted features and improve the correctness through probability tracking. The experimental result shows an average detection rate increase of 13.2% over existing works.

Fisheye lenses are widely used when the users want to capture the image/video with wide field of view (FoV) which is particularly suited to surveillance monitoring and vehicle camera. However, no projection from the actual scene in wide FoV image can avoid the distortion. If this problem cannot be solved, the fisheye image will difficult be used for object detection or analysis due to the distorted shapes of the scene objects. To correct this problem and obtain the natural-looking image, a two-step correction approach is proposed. Firstly, adaptive latitude and longitude correction are presented and the Hough transform is used to detect and estimate the straight-line. Secondly, the straight-line preserving and orientation, consistency based optimization is examined to obtain the final correction result. To compare the effectiveness of the proposed method, some fisheye correction methods are discussed. The experimental results demonstrate that the proposed method can obtain the coherent natural-looking.

Huge variety of medicine cures diseases. But unlabeled pills sometimes confuse people, even causing adverse drug events. This paper introduces a high accuracy automatic pill recognition method based on pill imprint which is a main discriminative factor between different pills. To describe the imprint information clearly, we propose a Two-step Sampling Distance Sets (TSDS) descriptor based on Distance Sets (DS) using a two-step sampling strategy. The two-step sampling strategy applies a resampling according to imprint segmentation, which divides an imprint into separated strokes, fragments and noise points. The TSDS is able to take control over the selection of feature points, aiming to cut down the noise points and unwished fragments generated by imprint extraction which will cause disturbance on recognition. In the aspect of the imprint extraction, we preprocess the pill image by dynamic contrast adjustment to cope with the exposure problem. Modified Stroke Width Transform (MSWT) is used to extract the imprint by detecting the coherent strokes on the pill. Finally, several experimental results have shown 86.01%, rank-1 matching accuracy, and 93.64%, within top 5 ranks, when classifying pills into 2500 categories.

In Kitakyushu city, more than 25 % people are older than 65 years old. The roads in residential area for the people are very narrow, very steep slope and vulnerable. Some needs assessment for small electric vehicle at some event and community activity was done. Based on the survey result and topographical features, the small electric vehicle for elderly person was selected and was modified and the automatic driving system was built. Automatic driving and platooning using Zigbee or Digimesh to exchange driving data between vehicles were built and tested. © 2014 IFAC.

This paper presents an essential algorithm for optimization-based image processing using the bilateral filter (BLF), called constant-time transposed BLF (O(1) TBLF). Some iterative solvers for optimization problems require a pair of filters defined as multiplying a filter matrix or its transpose to vectorized images. Since the BLF can be described as a matrix form, its paired filter also exists, called a TBLF in this paper. BLF-based optimization achieves high smoothing performance; whereas, it requires much high computational complexity due to iterating both BLF and TBLF many times. Hence, this paper designs an O(1) TBLF algorithm to accelerate the iterative process. Experiments show that our O(1) TBLF runs in low complexity regardless of its filter window size and works effectively for flash/no-flash image integration via BLF-based optimization.

In this paper, a novel discriminative dictionary learning with pairwise constraints by maximum correntropy criterion is proposed for pair matching problem. Comparing with the conventional dictionary learning approaches, the proposed method has several advantages: (i) It can deal with the outliers and noises problem more efficiently during the reconstruction step. (ii) It can be effectively solved by half-quadratic optimization algorithm, and in each iteration step, the complex optimization problem can be reduced to a general problem that can be efficiently solved by feature-sign search optimization. (iii) The proposed method is capable of analyzing non-Gaussian noise to reduce the influence of large outliers substantially, resulting in a robust and discriminative dictionary. We test the performance of the proposed method on two applications: face verification on the challenging restricted protocol of Labeled Faces in the Wild (LFW) benchmark and face-track identification on a dataset with more than 7,000 face-tracks. Compared with the recent state-of-the-art approaches, the outstanding performance of the proposed method validates its robustness and discriminability. © 2013 IEEE.

This paper proposes an improved color barycenter model (CBM) and its separation for automatic road sign (RS) detection. The previous version of CBM can find out the colors of RS, but the accuracy is not high enough for separating the magenta and blue regions and the influence of number with the same color are not considered. In this paper, the improved CBM expands the barycenter distribution to cylindrical coordinate system (CCS) and takes the number of colors at each position into account for clustering. Under this distribution, the color information can be represented more clearly for analyzing. Then aim to the characteristic of barycenter distribution in CBM (CBM-BD), a constrained clustering method is presented to cluster the CBM-BD in CCS. Although the proposed clustering method looks like conventional K-means in some part, it can solve some limitations of K-means in our research. The experimental results show that the proposed method is able to detect RS with high robustness.

This paper presents a novel color descriptor based on the proposed Color Barycenter Hexagon (CBH) model for automatic Road-Sign (RS) detection. In the visual Driver Assistance System (DAS), RS detection is one of the most important factors. The system provides drivers with important information on driving safety. Different color combinations of RS indicate different functionalities; hence a robust color detector should be designed to address color changes in natural surroundings. The CBH model is constructed with barycenter distribution in the created color triangle, which represents RS colors in a more compact way. For detecting RS, the CBH model is used to segment color information at the initial step. Furthermore, a judgment process is applied to verify each RS candidate through the size, aspect ratio, and color ratio. Experimental results show that the proposed method is able to detect RS with robust, accurate performance and is invariant to light and scale in more complex surroundings.

Linear Discriminant Analysis (LDA) is a well-known feature extraction method for supervised subspace learning in statistical pattern recognition. In this paper, a novel method of LDA based on a new L1-norm optimization technique and its variances are proposed. The conventional LDA, which is based on L2-norm, is sensitivity to the presence of outliers, since it used the L2-norm to measure the between-class and within-class distances. In addition, the conventional LDA often suffers from the so-called small sample size (3S) problem since the number of samples is always smaller than the dimension of the feature space in many applications, such as face recognition. Based on L1-norm, the proposed methods have several advantages, first they are robust to outliers because they utilize the L1-norm, which is less sensitive to outliers. Second, they have no 3S problem. Third, they are invariant to rotations as well. The proposed methods are capable of reducing the influence of outliers substantially, resulting in a robust classification. Performance assessment in face application shows that the proposed approaches are more effectiveness to address outliers issue than traditional ones.

This paper proposes an improved color barycenter model (CBM) for road sign detection. The previous version of CBM can find out the colors of road-sign (RS), but its accuracy is not high enough for magenta and blue region segmentation. The improved CBM extends the barycenter distribution to cylinder coordinate and takes the number of colors in every point into account. Then the K-meansclusteringisusedtoanalyze the distribution under cylinder coordinate. Using Geodesic distance instead of Euclidean distance for K-means clustering and some conditions provided by the initial color region of CBM is used to constrain K-means operation. The experimental results show that the improved method is able to detect RS with high robustness.

In this paper, the color barycenter model (CBM) based image enhancement method using multihistogram mapping and merging is presented. Generally, histogram analysis based methods are effective for contrast enhancement, but this kind of method is hard to enhance the dark and bright regions efficiently simultaneously, such as the back-light image. To solve this problem, a mapping function is studied for multihistogram mapping to obtain several images with different contrast, and merging them by the best patch selecting of every position. Firstly, using the CBM to calculate the gray component as the input data. Secondly, obtaining several image with different contrast by our mapping function. Thirdly, calculating the gradient feature of the separated patches and selecting the best ones for merging. Finally, using the mix Gaussian filter to smooth the merged image. Based on the proposed approach, enhancement can be achieved for global/local regions under different light conditions. The experimental results show better effectiveness than other methods.

In Kitakyushu city there are more than a quarter of people are older than 65-year-old. The road in residential area of this area has specific characters including very narrow road, very steep slope and vulnerable roadbed. Some needs assessment of small electric vehicle at some event activity and community was done. Based on the survey result and topographical features, the small electric vehicle for elder was designed and modified and automatic driving system was built. Automatic driving and platooning using Zigbee to exchange driving data between vehicles were built and tested.

Great achievements in modern medicine benefit human beings. Also, it has brought about an explosive growth of pharmaceuticals that current in the market. In daily life, pharmaceuticals sometimes confuse people when they are found unlabeled. In this paper, we propose an automatic pill recognition technique to solve this problem. It functions mainly based on the imprint feature of the pills, which is extracted by proposed MSWT (modified stroke width transform) and described by WSC (weighted shape context). Experiments show that our proposed pill recognition method can reach an accurate rate up to 92.03% within top 5 ranks when trying to classify more than 10 thousand query pill images into around 2000 categories. © 2013 SPIE.

In this paper, we propose Learned Local Curvelet Patterns (LLCP) for presenting the local features of facial images. The proposed method is based on curvelet transform which can overcome the weakness of traditional Gabor wavelets in higher dimension, and better capture the curve singularities and hyperplane singularities of facial images. Different from wavelet transform, curvelet transform can effectively and efficiently approximate the curved edges with very few coefficients as well as taking space-frequency information into consideration. First, LLCP designs several learned codebooks from Curvelet filtered facial images. Then each facial image can be encoded into multiple pattern maps and finally block-based histograms of these patterns are concatenated into an histogram sequence to be used as a face descriptor. In order to reduce the face feature descriptor, 2-Directional L1-Norm Based 2DPCA ((2D)2PCA-L1) is proposed which is simultaneously considering the row and column directions for efficient face representation and recognition. Performance assessment in several face recognition problem shows that the proposed approach is superior to traditional ones. © 2013 Springer-Verlag.

Linear Discriminant Analysis (LDA) is a famous supervised feature extraction method for subspace learning in computer vision and pattern recognition. In this paper, a novel method of LDA based on a new Maximum Correntropy Criterion optimization technique is proposed. The conventional LDA, which is based on L2-norm, is sensitivity to the presence of outliers. The proposed method has several advantages: first, it is robust to large outliers. Second, it is invariant to rotations. Third, it can be effectively solved by half-quadratic optimization algorithm. And in each iteration step, the complex optimization problem can be reduced to a quadratic problem that can be efficiently solved by a weighted eigenvalue optimization method. The proposed method is capable of analyzing non-Gaussian noise to reduce the influence of large outliers substantially, resulting in a robust classification. Performance assessment in several datasets shows that the proposed approach is more effectiveness to address outlier issue than traditional ones. © 2013 Springer-Verlag.

The topic of retrieving videos containing a desired person by using the content of faces without any help of textual information has many interesting applications like video surveillance, social network, video mining, etc. However, face-by-face matching leads to an unacceptable response time for a video dataset with a large number of detected faces and may also reduce the accuracy of searching. Therefore, in this paper we propose a scheme to generate facial signatures for fast retrieving videos containing the same person with a query. First, we summarize each video as a set of person-oriented individuals based on detected faces, which are represented as high dimensional vectors in a feature space. Then, each person with a collection of high dimensional vectors is projected to a compact and reduced dimensionality representation that is called facial signature for this person. The projection is realized by constructing a matcher using linear discriminant analysis with maximum correntropy criterion optimization. In this research, two kinds of signatures are provided, which are called 1D facial signature and 2D facial signature. The proposed searching scheme can support two types of queries: face image and video clip. Evaluations on a large dataset of videos show reliable measurement of similarities by using facial signature to represent each person generated from videos and also demonstrate that the proposed searching scheme has the potential to substantially reduce the response time and slightly increase the mean average precision of retrieval. © 2013 IEEE.

The topic of retrieving videos containing a desired person from a dataset just using the content of faces without any help of textual information has many interesting applications like video surveillance, social network, video mining, etc. However, traditional face matching against a huge number of detected faces leads to an unacceptable response time and may also reduce the accuracy due to the large variations in facial expressions, poses, lighting, etc. Therefore, in this paper we propose a novel method to generate discriminative "signatures" for efficiently retrieving the videos containing the same person with a query. In this research, the signature is defined as a compact, discriminative and reduced dimensionality representation, which is generated from a set of high-dimensional feature vectors of an individual. The desired videos are retrieved based on the similarities between the signature of the query and those of individuals in the database. In particular, we make the following contributions. Firstly, we give an algorithm of two directional linear discriminant analysis with maximum correntropy criterion (2DLDA-MCC) as an extension to our recently proposed maximum correntropy criterion based linear discriminant analysis (LDA-MCC). Both algorithms are robust to outliers and noise. Secondly, we present an approach for transferring a set of exemplars to a fixed-length signature using LDA-MCC and 2DLDA-MCC, resulting in two kinds of signatures that are called 1D signature and 2D signature. Finally, a novel video retrieval scheme is given based on the signatures, which has low storage requirement and can achieve a fast search. Evaluations on a large dataset of videos show reliable measurement of similarities by using the proposed signatures to represent the identities generated from videos. Experimental results also demonstrate that the proposed video retrieval scheme has the potential to substantially reduce the response time and slightly increase the mean average precision of retrieval. Copyright © 2013 The Institute of Electronics, Information and Communication Engineers.

The topic of retrieving videos containing a desired person by just using facial content has many applications like video surveillance, social network, etc. In this paper, we propose a compact, discriminative and low-dimensional signature to describe an person with a set of high-dimensional features. The signature is generated by linear discriminant analysis with maximum correntropy criterion that is robust to outliers and noises. Based on the proposed signatures, a new video retrieval scheme is given for fast finding the desired videos by measuring the similarities between the signature of a query and the ones in the dataset. Evaluations on a large dataset of videos show that the proposed video retrieval scheme has the potential to substantially reduce the response time and slightly increase the mean average precision of retrieval.

This paper presents a novel approach to remove the specular reflections on the transparent plastic medicine package and automatically extract the randomly distributed pills inside. In this approach, three cameras are employed to take images of the package from different viewpoints. And these three images are used as input image set while the output is a series of small images of a single pill. And these images can be directly applied to the traditional single pill recognition algorithms. The experimental results show the reliability of our approach by measuring correct detection rate (100%), false detection rate (0%) and pill separation accuracy (98.4%). And the proposed method processes a set of three 725 x 725 sized images at 0.15s averagely on a Core i5-2400 3.1GHz PC.

There has been significant recent interest in stereo correspondence algorithms for use in the urban automotive environment [1, 2, 3]. In this paper we evaluate a range of dense stereo algorithms, using a unique evaluation criterion which provides quantitative analysis of accuracy against range, based on ground truth 3D annotated object information. The results show that while some algorithms provide greater scene coverage, we see little differentiation in accuracy over short ranges, while the converse is shown over longer ranges. Within our long range accuracy analysis we see a distinct separation of relative algorithm performance. This study extends prior work on dense stereo evaluation of Block Matching (BM)[4], Semi-Global Block Matching (SGBM)[5], No Maximal Disparity (NoMD)[6], Cross[7], Adaptive Dynamic Programming (AdptDP)[8], Efficient Large Scale (ELAS)[9], Minimum Spanning Forest (MSF)[10] and Non-Local Aggregation (NLA)[11] using a novel quantitative metric relative to object range.

This paper presents an efficient constant-time Gaussian filter which provides a high accuracy at a low cost over a wide range of scale sigma. It requires only 14 multiplications per pixel in image filtering regardless of sigma, which is fewer than state-of-the- art constant-time Gaussian filters. Main ideas of the paper are as follows: 1) introducing a second-order shift property of the discrete cosine transform type-5 (DCT-5) to convolve cosines faster, and 2) suppressing error propagation caused by the shift property. Experiments in image processing show that the proposed algorithm is 3.7 x faster than a state-of-the-art recursive Gaussian filter and comparable to that of +/- 3 sigma-ssupported Gaussian convolution with sigma = 2.33. The output accuracy is stable at around 80 [dB] all over sigma is an element of [1, 128].

The topic of retrieving videos containing a desired person by using the content of faces without any help of textual information has many interesting applications like video surveillance, social network, video mining, etc. However, face-by-face matching leads to an unacceptable response time for a video dataset with a large number of detected faces and may also reduce the accuracy of searching. Therefore, in this paper we propose a scheme to generate facial signatures for fast retrieving videos containing the same person with a query. First, we summarize each video as a set of person-oriented individuals based on detected faces, which are represented as high dimensional vectors in a feature space. Then, each person with a collection of high dimensional vectors is projected to a compact and reduced dimensionality representation that is called facial signature for this person. The projection is realized by constructing a matcher using linear discriminant analysis with maximum correntropy criterion optimization. In this research, two kinds of signatures are provided, which are called 1D facial signature and 2D facial signature. The proposed searching scheme can support two types of queries: face image and video clip. Evaluations on a large dataset of videos show reliable measurement of similarities by using facial signature to represent each person generated from videos and also demonstrate that the proposed searching scheme has the potential to substantially reduce the response time and slightly increase the mean average precision of retrieval. © 2013 IEEE.

The topic of retrieving videos containing a desired person by just using facial content has many applications like video surveillance, social network, etc. In this paper, we propose a compact, discriminative and low-dimensional signature to describe an person with a set of high-dimensional features. The signature is generated by linear discriminant analysis with maximum correntropy criterion that is robust to outliers and noises. Based on the proposed signatures, a new video retrieval scheme is given for fast finding the desired videos by measuring the similarities between the signature of a query and the ones in the dataset. Evaluations on a large dataset of videos show that the proposed video retrieval scheme has the potential to substantially reduce the response time and slightly increase the mean average precision of retrieval.

This paper presents a novel approach to remove the specular reflections on the transparent plastic medicine package and automatically extract the randomly distributed pills inside. In this approach, three cameras are employed to take images of the package from different viewpoints. And these three images are used as input image set while the output is a series of small images of a single pill. And these images can be directly applied to the traditional single pill recognition algorithms. The experimental results show the reliability of our approach by measuring correct detection rate (100%), false detection rate (0%) and pill separation accuracy (98.4%). And the proposed method processes a set of three 725 x 725 sized images at 0.15s averagely on a Core i5-2400 3.1GHz PC.

This paper presents an efficient constant-time Gaussian filter which provides a high accuracy at a low cost over a wide range of scale sigma. It requires only 14 multiplications per pixel in image filtering regardless of sigma, which is fewer than state-of-the- art constant-time Gaussian filters. Main ideas of the paper are as follows: 1) introducing a second-order shift property of the discrete cosine transform type-5 (DCT-5) to convolve cosines faster, and 2) suppressing error propagation caused by the shift property. Experiments in image processing show that the proposed algorithm is 3.7 x faster than a state-of-the-art recursive Gaussian filter and comparable to that of +/- 3 sigma-ssupported Gaussian convolution with sigma = 2.33. The output accuracy is stable at around 80 [dB] all over sigma is an element of [1, 128].

In this paper, we propose Local Curve let Binary Patterns (LCBP) and Learned Local Curve let Patterns (LLCP) for presenting the local features of facial images. The proposed methods are based on Curve let transform which can overcome the weakness of traditional Gabor wavelets in higher dimensions, and better capture the curve singularities and hyperplane singularities of facial images. LCBP can be regarded as a combination of Curve let features and LBP operator while LLCP designs several learned codebooks from patch sets, which are constructed by sampling patches from Curvelet filtered facial images. Each facial image can be encoded into multiple pattern maps and block-based histograms of these patterns are concatenated into an histogram sequence to be used as a face descriptor. During the face representation phase, one input patch is encoded by one pattern in LCBP while multi-patterns in LLCP. Finally, an effective classifier called Weighted Histogram Spatially constrained Earth Mover's Distance (WHSEMD) which utilizes the discriminative powers of different facial parts, the different patterns and the spatial information of face is proposed. Performance assessment in face recognition and gender estimation under different challenges shows that the proposed approaches are superior than traditional ones.

In this paper, we propose a novel Hand Posture Recognition (HPR) for biometrics. This study uses the three dimensional point clouds for robust hand posture recognition at the rotation and scale. Multi-Hilbert Scanning Distance (MHSD) are also introduced for mathematical approaches of shape matching. HPR framework is divided into five parts: detecting hand region, removing the wrist, aligning the hand pose, extracting feature descriptor, and matching. Based on the experimental results, this framework showed superior results for hand posture recognition rate. © 2012 EURASIP.

In this paper, an adaptive framework based on histogram separation and mapping for image contrast enhancement is presented. In this framework, the histogram is separated by binary tree structure with the proposed adaptive histogram separation strategy. Generally, histogram equalization (HE) is an effective technique for contrast enhancement. However, the conventional HE usually gives the processed image with unnatural look and artifacts by excessive enhancement. For overcoming this shortage, the adaptive histogram separation unit (AHSU) is proposed to convert the global enhancement problem into local. And for mapping the histogram partitions into more optimal ranges, the exact histogram separation is discussed. Finally, an adaptive histogram separation and mapping framework (AHSMF) for contrast enhancement is presented, and the experimental results show better effectiveness than other histogram based methods.

With the wide usage of multispectral images, a fast efficient multidimensional clustering method becomes not only meaningful but also necessary. In general, to speed up the multidimensional images' analysis, a multidimensional feature vector should be transformed into a lower dimensional space. The Hilbert curve is a continuous one-to-one mapping from N-dimensional space to one-dimensional space, and can preserves neighborhood as much as possible. However, because the Hilbert curve is generated by a recurve division process, 'Boundary Effects' will happen, which means data that are close in N-dimensional space may not be close in one-dimensional Hilbert order. In this paper, a new efficient approach based on the space-filling curves is proposed for classifying multispectral satellite images. In order to remove 'Boundary Effects' of the Hilbert curve, multiple Hilbert curves, z curves, and the Pseudo-Hilbert curve are used jointly. The proposed method extracts category clusters from one-dimensional data without computing any distance in N-dimensional space. Furthermore, multispectral images can be analyzed hierarchically from coarse data distribution to fine data distribution in accordance with different application. The experimental results performed on LANDSAT data have demonstrated that the proposed method is efficient to manage the multispectral images and can be applied easily.

In this paper, we propose a novel gait recognition framework - Spherical Space Model with Human Point Clouds (SSM-HPC) to recognize front view of human gait. A new gait representation - Marching in Place (MIP) gait is also introduced which preserves the spatiotemporal characteristics of individual gait manner. In comparison with the previous studies on gait recognition which usually use human silhouette images from image sequences, this research applies three dimensional (3D) point clouds data of human body obtained from stereo camera. The proposed framework exhibits gait recognition rates superior to those of other gait recognition methods.

In this paper, we propose a segmentation based method to estimate the haze-free image by the optical model. In this work, we estimate the atmospheric light by color barycenter hexagon (CBH) model and use the watershed to segment the image to calculate transmission map by dark pixels with single image. Firstly, non-color region is segmented by CBH model and calculate the atmospheric light. Then, use the watershed with rang component of CBH model to segment the color image into several sub-regions, and estimate the transmission map. Finally, use the optical model with the parameters to restore the haze-free image. The experimental results show that our method is more effective and able to get better results than other compared single image based methods.

We are interested in retrieving video shots or videos containing particular people from a video dataset. Owing to the large variations in pose, illumination conditions, occlusions, hairstyles and facial expressions, face tracks have recently been researched in the fields of face recognition, face retrieval and name labeling from videos. However, when the number of face tracks is very large, conventional methods, which match all or some pairs of faces in face tracks, will not be effective. Therefore, in this paper, an efficient method for finding a given person from a video dataset is presented. In our study, in according to performing research on face tracks in a single video, we also consider how to organize all the faces in videos in a dataset and how to improve the search quality in the query process. Different videos may include the same person; thus, the management of individuals in different videos will be useful for their retrieval. The proposed method includes the following three points. (i) Face tracks of the same person appearing for a period in each video are first connected on the basis of scene information with a time constriction, then all the people in one video are organized by a proposed hierarchical clustering method. (ii) After obtaining the organizational structure of all the people in one video, the people are organized into an upper layer by affinity propagation. (iii) Finally, in the process of querying, a remeasuring method based on the index structure of videos is performed to improve the retrieval accuracy. We also build a video dataset that contains six types of videos: films, TV shows, educational videos, interviews, press conferences and domestic activities. The formation of face tracks in the six types of videos is first researched, then experiments are performed on this video dataset containing more than 1 million faces and 218,786 face tracks. The results show that the proposed approach has high search quality and a short search time.

Polar and Spherical Fourier analysis can be used to extract rotation invariant features for image retrieval and pattern recognition tasks. They are demonstrated to show superiorities comparing with other methods on describing rotation invariant features of two and three dimensional images. Based on mathematical properties of trigonometric functions and associated Legendre polynomials, fast algorithms are proposed for multimedia applications like real time systems and large multimedia databases in order to increase the computation speed. The symmetric points are computed simultaneously. Inspired by relative prime number theory, systematic analysis are given in this paper. Novel algorithm is deduced that provide even faster speed. Proposed method are 9-15% faster than previous work. The experimental results on two and three dimensional images are given to illustrate the effectiveness of the proposed method. Multimedia signal processing applications that need real time polar and spherical Fourier analysis can be benefit from this work.

In this paper, we propose a novel approach for presenting the local features of digital image using 1D Local Patterns by Multi-Scans (1DLPMS). We also consider the extentions and simplifications of the proposed approach into facial images analysis. The proposed approach consists of three steps. At the first step, the gray values of pixels in image are represented as a vector giving the local neighborhood intensity distrubutions of the pixels. Then, multi-scans are applied to capture different spatial information on the image with advantage of less computation than other traditional ways, such as Local Binary Patterns (LBP). The second step is encoding the local features based on different encoding rules using 1D local patterns. This transformation is expected to be less sensitive to illumination variations besides preserving the appearance of images embedded in the original gray scale. At the final step, Grouped 1D Local Patterns by Multi-Scans (G1DLPMS) is applied to make the proposed approach computationally simpler and easy to extend. Next, we further formulate boosted algorithm to extract the most discriminant local features. The evaluated results demonstrate that the proposed approach outperforms the conventional approaches in terms of accuracy in applications of face recognition, gender estimation and facial expression.

This paper presents a color-based method for medicine package recognition, called a linear manifold color descriptor (LMCD). It describes a color distribution (a set of color pixels) of a color package image as a linear manifold (an affine subspace) in the color space, and recognizes an anonymous package by linear manifold matching. Mainly due to low dimensionality of color spaces, LMCD can provide more compact description and faster computation than description styles based on histogram and dominant-color. This paper also proposes distance-based dissimilarities for linear manifold matching. Specially designed for color distribution matching, the proposed dissimilarities are theoretically appropriate more than J-divergence and canonical angles. Experiments on medicine package recognition validates that LMCD outperforms competitors including MPEG-7 color descriptors in terms of description size, computational cost and recognition rate.

The Hilbert curve is a one-to-one mapping between multidimensional space and one-dimensional (1-D) space. Due to the advantage of preserving high correlation of multidimensional points, it receives much attention in many areas. Especially in image processing, Hilbert curve is studied actively as a scan technique (Hilbert scan). Currently there have been several Hilbert scan algorithms, but they usually have strict implementation conditions. For example, they use recursive functions to generate scans, which makes the algorithms complex and difficult to implement in real-time systems. Moreover the length of each side in a scanned region should be same and equal to the power of two, which limits the application of Hilbert scan greatly. In this paper, to remove the constraints and improve the Hilbert scan for a general application, an effective generalized three-dimensional (3-D) Hilbert scan algorithm is proposed. The proposed algorithm uses two simple look-up tables instead of recursive functions to generate a scan, which greatly reduces the computational complexity and saves storage memory. Furthermore, the experimental results show that the proposed generalized Hilbert scan can also take advantage of the high correlation between neighboring lattice points in an arbitrarily-sized cuboid region, and give competitive performance in comparison with some common scan techniques. (C) 2012 Elsevier Inc. All rights reserved.

Hypercomplex polar Fourier analysis treats a signal as a vector field and generalizes the conventional polar Fourier analysis. It can handle signals represented by hypercomplex numbers such as color images. Hypercomplex polar Fourier analysis is reversible that means it can reconstruct image. Its coefficient has rotation invariance property that can be used for feature extraction. However in order to increase the computation speed, fast algorithm is needed especially for image processing applications like realtime systems and limited resource platforms. This paper presents fast hypercomplex polar Fourier analysis based on symmetric properties and mathematical properties of trigonometric functions. Proposed fast hypercomplex polar Fourier analysis computes symmetric points simultaneously, which significantly reduce the computation time.

In this paper, we propose a novel Hand Posture Recognition (HPR) for biometrics. This study uses the three dimensional point clouds for robust hand posture recognition at the rotation and scale. Multi-Hilbert Scanning Distance (MHSD) are also introduced for mathematical approaches of shape matching. HPR framework is divided into five parts: detecting hand region, removing the wrist, aligning the hand pose, extracting feature descriptor, and matching. Based on the experimental results, this framework showed superior results for hand posture recognition rate.

This paper presents a color-based method for medicine package recognition, called a linear manifold color descriptor (LMCD). It describes a color distribution (a set of color pixels) of a color package image as a linear manifold (an affine subspace) in the color space, and recognizes an anonymous package by linear manifold matching. Mainly due to low dimensionality of color spaces, LMCD can provide more compact description and faster computation than description styles based on histogram and dominant-color. This paper also proposes distance-based dissimilarities for linear manifold matching. Specially designed for color distribution matching, the proposed dissimilarities are theoretically appropriate more than J-divergence and canonical angles. Experiments on medicine package recognition validates that LMCD outperforms competitors including MPEG-7 color descriptors in terms of description size, computational cost and recognition rate. Copyright © 2012 The Institute of Electronics, Information and Communication Engineers.

Polar and Spherical Fourier analysis can be used to extract rotation invariant features for image retrieval and pattern recognition tasks. They are demonstrated to show superiorities comparing with other methods on describing rotation invariant features of two and three dimensional images. Based on mathematical properties of trigonometric functions and associated Legendre polynomials, fast algorithms are proposed for multimedia applications like real time systems and large multimedia databases in order to increase the computation speed. The symmetric points are computed simultaneously. Inspired by relative prime number theory, systematic analysis are given in this paper. Novel algorithm is deduced that provide even faster speed. Proposed method are 9-15% faster than previous work. The experimental results on two and three dimensional images are given to illustrate the effectiveness of the proposed method. Multimedia signal processing applications that need real time polar and spherical Fourier analysis can be benefit from this work. Copyright © 2012 The Institute of Electronics, Information and Communication Engineers.

In this paper, we propose a novel approach for presenting the local features of digital image using 1D Local Patterns by Multi-Scans (1DLPMS). We also consider the extentions and simplifications of the proposed approach into facial images analysis. The proposed approach consists of three steps. At the first step, the gray values of pixels in image are represented as a vector giving the local neighborhood intensity distrubutions of the pixels. Then, multi-scans are applied to capture different spatial information on the image with advantage of less computation than other traditional ways, such as Local Binary Patterns (LBP). The second step is encoding the local features based on different encoding rules using 1D local patterns. This transformation is expected to be less sensitive to illumination variations besides preserving the appearance of images embedded in the original gray scale. At the final step, Grouped 1D Local Patterns by Multi-Scans (G1DLPMS) is applied to make the proposed approach computationally simpler and easy to extend. Next, we further formulate boosted algorithm to extract the most discriminant local features. The evaluated results demonstrate that the proposed approach outperforms the conventional approaches in terms of accuracy in applications of face recognition, gender estimation and facial expression. Copyright © 2012 The Institute of Electronics, Information and Communication Engineers.

In this paper, we propose a novel gait recognition framework - Spherical Space Model with Human Point Clouds (SSM-HPC) to recognize front view of human gait. A new gait representation - Marching in Place (MIP) gait is also introduced which preserves the spatiotemporal characteristics of individual gait manner. In comparison with the previous studies on gait recognition which usually use human silhouette images from image sequences, this research applies three dimensional (3D) point clouds data of human body obtained from stereo camera. The proposed framework exhibits gait recognition rates superior to those of other gait recognition methods. © 2012 The Institute of Electronics, Information and Communication Engineers.

This paper presents an approximate Gaussian filter which can run in one-pass with high accuracy based on spectrum sparsity. This method is a modification of the cosine integral image (CII), which decomposes a filter kernel into few cosine terms and convolves each cosine term with an input image in constant time per pixel by using integral images and look-up tables. However, they require much workspace and high access cost. The proposed method solves the problem with no decline in quality by sequentially updating sums instead of integral images and by improving look-up tables, which accomplishes a one-pass approximation with much less workspace. A specialization for tiny kernels are also discussed for faster calculation. Experiments on image filtering show that the proposed method can run nearly two times faster than CII and also than convolution even with small kernel.

We present a novel method to remove the specular reflections on the surface of transparent plastic film. Our approach uses four light sources with strategic positions to get four images. Based on the information that both reflection and shadow move a lot from image to image, we reconstruct a high quality image free from reflection and shadow by using a image set which is consist of four images. © 2012 IEEE.

In this paper, we put forward an interest point detection framework and combine with a spin image algorithm to register them. A framework is presented in this study. This method make use of the Harris Operator Extension method of interest point detection on 3D manifold triangular meshes in barycentric coordinate. Using this approach, we can extract the object correctly and effectively in noise situation. The unique advantage of this framework is its applicability to triangular meshes models. Experimental results on a different number of models are shown to demonstrate more accurate and effectively results for global registering 3D Objects triangular meshes for three pairs of corresponding interest point features.

The main purpose of interest points detection (IPD) for 3D objects based on Harris operator is to find the fast computation of weighted average of the derivative data for different points. In this paper, we analyze the extension of Harris operator using Hausdorff distance (EHOHD) and propose the extension of Harris operator using Hilbert scanning distance (EHOHSD) as a new proposed method to IPD on 3D manifold triangular meshes data. Proposed EHOHSD method is 6-16 times faster than EHOHD method. The quality of this IPD with EHOHSD was measured using the repeatability criterion.

In this paper, we present an efficient window-based stereo matching algorithm that especially focuses on foreground objects. For decades, there are a lot of researches about the stereo matching algorithms. However, most of methods concentrate on the entire pixels, which are time consuming and meaningless in the real applications. To strength the accuracy of stereo correspondence in foreground objects, a simple locally support-weight method based on the selected prime key is proposed in our algorithm. Moreover, a background pre-detection method is also employed to get a primary background checking map, which is used to reduce the number of computed pixels in the disparity selection. After the refinement of both foreground disparity map and background checking map, our algorithm obtains accurate disparity results on the foreground and separate it with the background by the correspondence search simultaneously. The experimental results based on the Middlebury stereo datasets demonstrate that our method can achieve a better performance on foreground disparity computing than many other support-weight methods in terms of both accuracy and computational efficiency. In addition, our proposals can make foreground objects detection easier at the same time.

This paper presents an approximate Gaussian filter which can run in one-pass with high accuracy based on spectrum sparsity. This method is a modification of the cosine integral image (CII), which decomposes a filter kernel into few cosine terms and convolves each cosine term with an input image in constant time per pixel by using integral images and look-up tables. However, they require much workspace and high access cost. The proposed method solves the problem with no decline in quality by sequentially updating sums instead of integral images and by improving look-up tables, which accomplishes a one-pass approximation with much less workspace. A specialization for tiny kernels are also discussed for faster calculation. Experiments on image filtering show that the proposed method can run nearly two times faster than CII and also than convolution even with small kernel.

In this paper, a method named histogram intersection metric learning from scene tracks is proposed for automatic organizing people in videos. We make the following contributions: (i) learning histogram intersection distance instead of Mahalanobis distance for widely used face features; (ii) learning the metric from scene tracks without manually labeling any examples, which enables learning across large variations in pose, expression, occlusion and illumination with small number of face pairs and can distinguish different people powerfully. We firstly test face identification, track clustering, and people organization on a long film, then individual retrieval based on people organization from a large video dataset is evaluated, demonstrating significantly increased search quality with respect to previous approaches on this area.

In this paper, we propose Local Curvelet Binary Patterns (LCBP) and Learned Local Curvelet Patterns (LLCP) for presenting the local features of facial images. The proposed methods are based on Curvelet transform which can overcome the weakness of traditional Gabor wavelets in higher dimensions, and better capture the curve singularities and hyperplane singularities of facial images. LCBP can be regarded as a combination of Curvelet features and LBP operator while LLCP designs several learned codebooks from patch sets, which are constructed by sampling patches from Curvelet filtered facial images. Each facial image can be encoded into multiple pattern maps and block-based histograms of these patterns are concatenated into an histogram sequence to be used as a face descriptor. During the face representation phase, one input patch is encoded by one pattern in LCBP while multi-patterns in LLCP. Finally, an effective classifier called Weighted Histogram Spatially constrained Earth Mover's Distance (WHSEMD) which utilizes the discriminative powers of different facial parts, the different patterns and the spatial information of face is proposed. Performance assessment in face recognition and gender estimation under different challenges shows that the proposed approaches are superior than traditional ones. Copyright © 2012 The Institute of Electronics, Information and Communication Engineers.

Polar Cosine Transform (PCT) is one of the Polar Harmonic Transforms that those kernels are basic waves and harmonic in nature. They are proposed to represent invariant patterns for two dimensional image description and are demonstrated to show superiorities comparing with other methods on extracting rotation invariant patterns for images. However in order to increase the computation speed, fast algorithm for PCT is proposed for real world applications like limited computing environments, large image databases and realtime systems. Based on our previous work, this paper novelly employs relative prime number theory to develop Fast Polar Cosine Transform (FPCT). The proposed FPCT is averagely over 11 ∼ 12.5 times faster than PCT that significantly boost computation process. The experimental results are given to illustrate the effectiveness of the proposed method.

This paper presents a color matching technique using weighted subspace on medicine package recognition. The proposed method is more compact and lowercomplex than scalable color descriptor and dominant color descriptor, which are employed by MPEG-7. Our method is based on subspace matching: A color object is treated as a subspace derived from its color distribution. Unlike mutual subspace method, it is specially designed for color matching. Specifically, weighted subspace and a distance-based dissimilarity are employed instead of normalized subspace and similarity based on canonical angles of MSM. Experiments show that the proposed method outperforms the conventional methods in terms of description size, building/matching speed, and recognition rate.

In this paper, a multi-feature selection and application based manifold learning metric method is proposed for Road-Sign Recognition (RSR). Firstly, the manifold metric between manifold from subspace is discussed in detail. After that, the multi-feature analyzing, selection, classification and application are introduced for rough recognition and create the manifold. Then the proposed method is used to evaluate the distance between the manifolds. Finally, the RSR results suggest that the proposed method is robust than other methods. © 2011 SICE.

Fourier transform is a significant tool in image processing and pattern recognition. By introducing a hypercomplex number, hypercomplex Fourier transform treats a signal as a vector field and generalizes the conventional Fourier transform. Inspired from that, hypercomplex polar Fourier analysis that extends conventional polar Fourier analysis is proposed in this paper. The proposed method can handle signals represented by hypercomplex numbers as color images. The hypercomplex polar Fourier analysis is reversible that means it can be used to reconstruct image. The hypercomplex polar Fourier descriptor has rotation invariance property that can be used for feature extraction. Due to the noncommutative property of quaternion multiplication, both left-side and right-side hypercomplex polar Fourier analysis are discussed and their relationships are also established in this paper. The experimental results on image reconstruction, rotation invariance, color plate test and image retrieval are given to illustrate the usefulness of the proposed method as an image analysis tool.

Generally, two problems of bag-of-features in image retrieval are still considered unsolved: one is that spatial information about descriptors is not employed well, which affects the accuracy of retrieval; the other is that the trade-off between vocabulary size and good precision, which decides the storage and retrieval performance. In this paper, we propose a novel approach called Hilbert scan based bag-of-features (HS-BoF) for image retrieval. Firstly, Hilbert scan based tree representation (HSBT) is studied, which is built based on the local descriptors while spatial relationships are added into the nodes by a novel grouping rule, resulting of a tree structure for each image. Further, we give two ways of codebook production based on HSBT: multi-layer codebook and multi-size codebook. Owing to the properties of Hilbert scanning and the merits of our grouping method, sub-regions of the tree are not only flexible to the distribution of local patches but also have hierarchical relations. Extensive experiments on caltech-256, 13-scene and 1 million Image Net images show that HS-BoF obtains higher accuracy with less memory usage.

Fourier transform is a significant tool in image processing and pattern recognition. By introducing hyper complex number, hyper complex Fourier transform [1] treats signal as vector field and generalizes conventional Fourier transform. Inspired from that, hypercomplex polar Fourier analysis is proposed in this paper. This work extends conventional polar Fourier analysis [5]. The proposed method can handle hypercomplex number represented signals like color image. The hypercomplex polar Fourier analysis is reversible that means it can be used to reconstruct image. The hypercomplex polar Fourier descriptor has rotation invariance property that can be used for feature extraction. Due to the noncommutative property of quaternion multiplication, both left-side and right-side hypercomplex polar Fourier analysis are discussed and their relationships are also established in this paper. The experimental results on image reconstruction, rotation invariance and color plate test are given to illustrate the usefulness of the proposed method as an image analysis tool.

In this paper, we propose a novel gait recognition framework which is Spherical Space Model with Human Point Clouds (SSM-HPC). A new gait representation is also introduced, which is called Marching in Place (MIP) gait and preserves the spatiotemporal characteristics of individual gait manner. Various researches for gait recognition have used human silhouette images from moving picture. This research uses Three Dimensional (3D) point clouds data of human body obtained from stereo camera, which has the scale-invariant property. The framework is applied for frontal view gait recognition. This framework showed superior results for gait recognition rate than other gait recognition methods.

This paper presents a low-level color descriptor which describes the color distribution of a color image as a weighted subspace in the color space, namely eigenvectors and eigen-values of the distribution. Thanks to low-dimensionality of color space, the proposed descriptor can provide compact description and fast computation. Furthermore, specialized for color distribution matching, it is more efficient than mutual subspace method (MSM). Experiments on medicine package recognition validate that the proposed descriptor outperforms MSM and MPEG-7 low-level color descriptors in terms of description size, computational cost and recognition rate.

In this paper, we propose a novel method to estimate the haze-free image based on the improved optical model. In this work, we calculate the objective transmission and distance transmission to estimate the haze-free image by the segmented hazy image. Firstly, the color clustering method is used to segment the image into several regions by color similarity for getting the objective transmission. Then, the graph-based segmentation is used to calculate the depth information for getting the distance transmission. Next, the atmosphere light is estimated according to the distance transmission Finally, the improved optical model is used to estimate the haze-free image. The experimental results show that our method is more effective and able to get better results than other single image based methods.

In image processing, anisotropic diffusion provides a forward method to remove noise while preserving edges accurate and sharp. However, due to the inappropriate edge estimation by gradient, some isolated noise points still exist and edge location is inaccurate. In this representation, isolated noise points and edges are distinguished by the significant difference of their "lengths", which are computed by orthogonally projecting their pixels to the corresponding normalized gradient directions and recording the number of the same projections. Combining gradient and "length" to estimate edges, isolated noise points are further suppressed while edges are re-located and enhanced. © 2011 IEEE.

Hypercomplex polar Fourier analysis treats a signal as a vector field and generalizes the conventional polar Fourier analysis. It can handle signals represented by hypercomplex numbers such as color images. It is reversible that can reconstruct image. Its coefficient has rotation invariance property that can be used for feature extraction. With these properties, it can be used for image processing applications like image representation and image understanding. However in order to increase the computation speed, fast algorithm is needed especially for image processing applications like realtime systems and limited resource platforms. This paper presents fast hypercomplex polar Fourier analysis that based on symmetric properties and mathematical properties of trigonometric functions. Proposed fast hypercomplex polar Fourier analysis computes symmetric eight points simultaneously that significantly reduce the computation time.

Hypercomplex polar Fourier analysis treats a signal as a vector field and generalizes the conventional polar Fourier analysis. It can handle signals represented by hypercomplex numbers such as color images. It is reversible that can reconstruct image. Its coefficient has rotation invariance property that can be used for feature extraction. With these properties, it can be used for image processing applications like image representation and image understanding. However in order to increase the computation speed, fast algorithm is needed especially for image processing applications like realtime systems and limited resource platforms. This paper presents fast hypercomplex polar Fourier analysis that based on symmetric properties and mathematical properties of trigonometric functions. Proposed fast hypercomplex polar Fourier analysis computes symmetric eight points simultaneously that significantly reduce the computation time.

Fourier transform is a significant tool in image processing and pattern recognition. By introducing hypercomplex number, hypercomplex Fourier transform [1] treats signal as vector field and generalizes conventional Fourier transform. Inspired from that, hypercomplex polar Fourier analysis is proposed in this paper. This work extends conventional polar Fourier analysis [5]. The proposed method can handle hypercomplex number represented signals like color image. The hypercom-plex polar Fourier analysis is reversible that means it can be used to reconstruct image. The hypercomplex polar Fourier descriptor has rotation invariance property that can be used for feature extraction. Due to the noncommutative property of quaternion multiplication, both left-side and right-side hypercomplex polar Fourier analysis are discussed and their relationships are also established in this paper. The experimental results on image reconstruction, rotation invariance and color plate test are given to illustrate the usefulness of the proposed method as an image analysis tool. © 2011 IEEE.

We are interested here in retrieving images containing a specific person in image database. Due to large variations in illumination conditions, hairstyles, facial expressions, etc. and the factors like occlusion, sunglasses, profile, etc., robust face matching has been a challenging problem. On the other hand, the speed of search is also a considerable issue, especially for the dataset with millions of face images. Inspired by face tracks in video retrieval which take advantages from the abundance of frames to get multiple exemplars, we present an approach named multi balanced trees for face retrieval from image dataset in this paper. Face images in the dataset are efficiently organized by the trees produced for persons. Multi sampling on the facial components employs the rich local information, which can help to differentiate different persons. Given a query face, a sorted face set with similarities is obtained by inserting the query into a tree. It is easy and fast to get the search results in respect that it avoids calculating the distances between query and elements in the cluster. In addition, a rectification strategy is given in the query process to rectify the error occurred in the generation of trees, resulting in a significant improvement of retrieval quality. Experimental results show the better face grouping ability in comparison with traditional methods. The speed of searching is improved as well. © 2011 IEEE.

There are a lot of previous works on common key encryptions such as DES, AES, etc, In this paper, a new common key encryption algorithm is proposed using Hilbert curves which are a one-to-one mapping between N-dimensional (N-D) spaces and 1-D space (a line). This is based on a property having a sharp rise in the number of Hilbert curve patterns in N-D spaces. In the case of N = 2, there are only four patterns, while if N is 5, the number of the patterns is more than 1 billions. Operations of addition and multiplication are denned on a curve, based on a mapping of a point in N-D spaces to a point on a line. In order to realize a cryptosystem, the algorithm utilizes Hilbert ordered point addresses, which is expressed as the coordinates of the points in N-dimensional space. © 2011 IEEE.

In this paper, we propose a novel gait recognition framework which is Spherical Space Model with Human Point Clouds (SSM-HPC). A new gait representation is also introduced, which is called Marching in Place (MIP) gait and preserves the spatiotemporal characteristics of individual gait manner. Various researches for gait recognition have used human silhouette images from moving picture. This research uses Three Dimensional (3D) point clouds data of human body obtained from stereo camera, which has the scale-invariant property. The framework is applied for frontal view gait recognition. This framework showed superior results for gait recognition rate than other gait recognition methods.

This paper presents a low-level color descriptor which describes the color distribution of a color image as a weighted subspace in the color space, namely eigenvectors and eigen-values of the distribution. Thanks to low-dimensionality of color space, the proposed descriptor can provide compact description and fast computation. Furthermore, specialized for color distribution matching, it is more efficient than mutual subspace method (MSM). Experiments on medicine package recognition validate that the proposed descriptor outperforms MSM and MPEG-7 low-level color descriptors in terms of description size, computational cost and recognition rate.

This paper presents an improved method for robust face recognition using illumination normalization based on Discrete Cosine Transform (DCT) in logarithm domain. Two novel coefficients are designed to identify the lighting condition (LC), based on which the low-frequency DCT coefficients are adaptively rescaled except the first one (DC). As a result variations under different illumination conditions are minimized meanwhile original information contained in low-frequency is comparatively well preserved. Results of experiments on Yale B database and Extended Yale B database show that proposed method has better performance under variational input illumination conditions. The proposed method is fast in computation and could be easily implemented into real time face recognition systems. © 2011 Springer-Verlag.

The influence of noise is an important problem on image acquisition and transmission stages. The traditional image denoising approaches only analyzing the pixels of local region with a moving window, which calculated by neighbor pixels to denoise. Recently, this research has been focused on the transform domain and feature space. Compare with the traditional approaches, the global multi-scale analyzing and unchangeable noise distribution is the advantage. Apparently, the estimation based methods can be used in transform domain and get better effect. This paper proposed a new approach to image denoising in orthonormal wavelet domain. In this paper, we adopt Stein's unbiased risk estimate (SURE) based method to denoise the low-frequency bands and the feature patches distance constraint (FPDC) method also be proposed to estimate the noise free bands in Wavelet domain. The key point is that how to divide the lower frequency sub-bands and the higher frequency sub-bands, and do interscale SURE and intrascale FPDC, respectively. We compared our denoising method with some well-known and new denoising algorithms, the experimental results show that the proposed method can give better performance and keep more detail information in most objective and subjective criteria than other methods.

Image denoising is a lively research field now. For solving this problem, non-linear filters based methods are the classical approach. These methods are based on local analysis of pixels with a moving window in spatial domain, but also have some shortcoming. Recently, because of the properties of Wavelet transform, this research has been focused on the wavelet domain. Compared to the classical nonlinear filters, the global multi-scale analysis characteristic of Wavelet is better for image denoising. So this paper proposed a new approach to use orthonormal Wavelet transform and distance constraint to solve this. Here, by minimizing the Stein's unbiased risk estimate (SURE) method to calculate the low frequency sub-band images for estimating. And convert the high frequency sub-band images to feature space, then use distance constraint to denoise by trained samples set. The experimental results show that the proposed method is efficiency and keep the detail ideally.

Polar Fourier Descriptor(PFD) and Spherical Fourier Descriptor(SFD) are rotation invariant feature descriptors for two dimensional(2D) and three dimensional(3D) image retrieval and pattern recognition tasks. They are demonstrated to show superiorities compared with other methods on describing rotation invariant features of 2D and 3D images. However in order to increase the computation speed, fast computation method is needed especially for machine vision applications like realtime systems, limited computing environments and large image databases. This paper presents fast computation method for PFD and SFD that are deduced based on mathematical properties of trigonometric functions and associated Legendre polynomials. Proposed fast PFD and SFD are 8 and 16 times faster than direct calculation that significantly boost computation process. Furthermore, the proposed methods are also compact for memory requirements for storing PFD and SFD basis in lookup tables. The experimental results on both synthetic and real data are given to illustrate the efficiency of the proposed method.

A common task of tone mapping algorithms is to reproduce high dynamic range images (HDR) on low dynamic range (LDR) display devices such as printers and monitors. We present a new tone mapping algorithm for the display of HDR images that was inspired by the adaptive process of the human visual system. The proposed algorithm is based on center/surround Retinex processing. Our method has two novel aspects. The input luminance image is first compressed by a global tone mapping curve. The curvature of the compression curve is adapted locally based on the pseudo-Hilbert scan technique, so it can provide a better overall impression before the subsequent local processing. Second, the local details are enhanced according to a non-linear adaptive spatial filter (Gaussian filter), whose shape (filter variance) is adapted to the high-contrast edges of the image. The proposed method takes advantage of the properties of both global and local processing while overcoming their respective disadvantages. Therefore, the algorithm can preserve visibility and contrast impression of high dynamic range scenes in standard display devices. We tested the proposed method on a variety of HDR images and also compared it to previous research. The results indicated that our method was effective for displaying images with high visual quality.

Road signs play an important role in our daily life which used to guide drivers to notice variety of road conditions and cautions. They provide important visual information that can help drivers operating their vehicles in a manner for enhancing traffic safety. The occurrence of some accidents can be reduced by using automatic road signs recognition system which can alert the drivers. This research attempts to develop a warning system to alert the drivers to notice the important road signs early enough to refrain road accidents from happening. For solving this, a non-linear weighted color enhancement method by pixels is presented. Due to the advantage of proposed method, different road signs can be detected from videos effectively. With suitably coefficients and operations, the experimental results have proved that the proposed method is robust, accurate and powerful in road signs detection.

Signature verification is a challenging task, because only a small set of genuine samples can be acquired and usually no forgeries are available in real application. In this paper, we propose a novel approach based on Hilbert scanning patterns and Gaussian mixture models for automatic on-line signature verification. Our system is composed of a similarity measure based on Hilbert scanning patterns and a simplified Gaussian mixture model for decision-level evaluation. To be practical, we introduce specific simplification ways for constructing a model and its training method. The system is compared to other state-of-the-art systems based on the results of the First International Signature Verification Competition (SVC 2004). Experiments are conducted to verify the effectiveness of our system. © 2010, The Institute of Image Electronics Engineers of Japan. All rights reserved.

Polar Fourier Descriptor(PFD) and Spherical Fourier Descriptor( SFD) are rotation invariant feature descriptors for two dimensional( 2D) and three dimensional(3D) image retrieval and pattern recognition tasks. They are demonstrated to show superiorities compared with other methods on describing rotation invariant features of 2D and 3D images. However in order to increase the computation speed, fast computation method is needed especially for machine vision applications like realtime systems, limited computing environments and large image databases. This paper presents fast computation method for PFD and SFD that are deduced based on mathematical properties of trigonometric functions and associated Legendre polynomials. Proposed fast PFD and SFD are 8 and 16 times faster than direct calculation that significantly boost computation process. Furthermore, the proposed methods are also compact for memory requirements for storing PFD and SFD basis in lookup tables. The experimental results on both synthetic and real data are given to illustrate the efficiency of the proposed method. Copyright © 2010 The Institute of Electronics, Information and Communication Engineers.

Polar Fourier Descriptor(PFD) and Spherical Fourier Descriptor(SFD) are rotation invariant feature descriptors for two dimensional(2D) and three dimensional(3D) image retrieval and pattern recognition tasks. They are demonstrated to show superiorities compared with other methods on describing rotation invariant features of 2D and 3D images. However in order to increase the computation speed, fast computation method is needed especially for applications like realtime systems and large image databases. This paper presents fast computation method for PFD and SFD that based on mathematical properties of trigonometric functions and associated Legendre polynomials. Proposed fast PFD and SFD are 8 and 16 times faster than traditional ones that significantly boost computation process. © 2010 IEEE.

This paper describes a novel method to match objects in cluttered scenes. This method makes use of Hilbert scanning of feature points in Hough space. We use a 3D Hough transform to obtain a spectrum on which 3D features are concentrated on the sphere. Then, based on the obtained Hough Spectrum, we apply Hilbert scanning on the sphere to match the objects. Using this approach, we can match the object correctly and robustly in both overlapping and noise situation. The characteristic of this method is that it is a global matching method without an estimate of the rotation first and suffering from computational complexity brought by voting/correlation procedure. The experiment results show that the method is more effective compared to existing methods in both matching rate and robustness.

In this work, a novel approach which called 1D local patterns by multi-scans(1DLPMS) for presenting the local features is proposed and its simplifications and extensions to facial image analysis are also considered. First, multi-scans are applied to capture different spatial information on the image with less computation than some traditional ways, such as Local Binary Patterns(LBP). Then, some 1D local patterns are given to encode the local features based on different coding rules. To make the proposed approach computationally simpler and easy to extend, grouped 1D local patterns by multiscans(G1DLPMS) is studied, which divides 1DLPMS into several groups and uses the co-occurrences of these groups. Performance assessment in face recognition under different challenges shows that the proposed approach is superior than traditional ones.

In this paper an efficient approach of document layout analysis and reading order determination is proposed for a reading robot. Firstly the input document images are preprocessed to remove noises, connect lines and domains, and to reduce the computation time. Secondly a bottom-up, parameter-independent, two-step layout analysis algorithm based on morphology is used, which outlines the geometry of the maximum homogeneous regions and classifies them into texts, tables, and pictures. Finally the reading order is determined, by a top-down recursive hierarchy algorithm derived from XY-cut, using a set of rules depending on layout information. Important parameters are acquired using statistic information of the given images to adapt to different types of documents. The proposed algorithm is applied to a large number of document images and the experimental results show that it makes the reading robot be able to read paper documents of different languages, even with complex layout structure.

This paper proposes an automatic 3D reconstruction approach for a Chinese talking face by a generic model and a single image. Firstly, an improved color-based ASM method is used to detect the face area and get the 2D face feature points automatically from the given image, which is not restricted to full frontal one. Then, color information is used to correct the location of face feature points. Finally, after text mapping, a particular and realistic 3D face model is deformed from a generic model. Using ASM face feature points extraction and correction based on skin color model, the problem of side face information missing is successfully resolved. Depending on only one image and one generic model, the computing cost of memory and time is largely reduced. The 3D face reconstructed can be easily deformed to form different expressions and mouth shapes. Experiments show that this approach is fast and efficient and has an output of a lifelike Chinese talking face.

Face finding is a very important initial step towards building up a fully automated face recognition system. Face detection by detecting skin like colors can achieve a high detection rate. In this paper, we presented a novel algorithm for face detection in color images with complex backgrounds. First a parallel structure for skin color detection is proposed to improve the accuracy of detections. The concept of the probability image has been introduced to utilize the color information in the traditional face detection methods specific for gray-scale images. After that, a classifier obtained from Adaboost training is applied to the result of skin detection to reduce the false positives. An experiment has been implemented to verify the improvement of this proposed research. And the proposed approach achieved a better result in this experiment.

Though elastic bunch graph matching (EBGM) has a good performance on face recognition in the distortion of facial expression, it is still not robust enough to in-depth rotation. To solve this problem, a novel face representation approach based on the space-filling tree is proposed in this paper. This kind of representation shows a better performance than Elastic bunch graph matching (EBGM) in in-depth rotation of pose especially when there are only frontal images in the training set. With the proposed face representation approach, the face recognition system is built. Experimental results on the FERET standard database show that the proposed face representation approach is more effective and robust to the in-depth rotation of pose when there are only frontal images in the training set.

In this paper, Hilbert scan based tree representation (HSBT) is presented for image search. Unlike common ways decreasing the number of interest points or reducing the dimensions of features or using searching methods to match interest points, the proposed method builds a tree for each image and gives a new distance measure to calculate the similarity between the query and images in database. In the proposed approach, Hilbert scan for arbitrarily-sized arrays is used to map the interest points from two-dimensional space to one-dimensional space at first. Then, interest points set is divided into several parts by a separation way, and a grouping strategy is given to build a tree for each image. Experimental results show that the proposed approach is space saving. That is because it only stores clustering center and relevant information of each node in the tree. It is also time saving since the similarity calculation is up to the nodes of tree rather than all the descriptors of image. At the same time, the retrieval precision is good, because Hilbert scanning preserves the correlation in two-dimensional image, so nodes of tree are shaped according to the compactness of interest points which can employ the local information as much as possible.

Polar Harmonic Transform (PHT) is termed to represent a set of transforms those kernels are basic waves and harmonic in nature. PHTs consist of Polar Complex Exponential Transform (PCET), Polar Cosine Transform (PCT) and Polar Sine Transform (PST). They are proposed to represent invariant image patterns for two dimensional image retrieval and pattern recognition tasks. They are demonstrated to show superiorities comparing with other methods on describing rotation invariant patterns for images. Kernel computation of PHTs is also simple and has no numerical stability issue. However in order to increase the computation speed, fast computation method is needed especially for real world applications like limited computing environments, large image databases and realtime systems. This paper presents Fast Polar Harmonic Transforms (FPHTs) including Fast Polar Complex Exponential Transform (FPCET), Fast Polar Cosine Transform (FPCT) and Fast Polar Sine Transform (FPST) that are deduced based on mathematical properties of trigonometric functions. The proposed FPHTs are averagely over 6 similar to 8 times faster than PHTs that significantly boost computation process. The experimental results on both synthetic and real data are given to illustrate the effectiveness of the proposed fast transforms.

Though elastic bunch graph matching (EBGM) has a good performance on face recognition in the distortion of facial expression, it is still not robust enough to in-depth rotation. To solve this problem, a novel face representation approach based on the space-filling tree is proposed in this paper. This kind of representation shows a better performance than Elastic bunch graph matching (EBGM) in in-depth rotation of pose especially when there are only frontal images in the training set. With the proposed face representation approach, the face recognition system is built. Experimental results on the FERET standard database show that the proposed face representation approach is more effective and robust to the in-depth rotation of pose when there are only frontal images in the training set.

In this work, we present a novel local pattern descriptor, Local Gradient Derivative Pattern (LGDP) to face recognition which considers more detailed information than the Local Binary Pattern (LBP). The face image is first divided into several small regions from which Local Gradient Derivative Pattern (LGDP) histograms are extracted and concatenated into a single, spatially enhanced feature vector to be used as a face descriptor. Three well-known and challenge-ORL, Yale and FERET face databases are used in the performances to evaluate the method. The experiments result clearly show that the proposed method give us a better performance than some other methods.

This paper proposes an automatic 3D reconstruction approach for a Chinese talking face by a generic model and a single image. Firstly, an improved color-based ASM method is used to detect the face area and get the 2D face feature points automatically from the given image, which is not restricted to full frontal one. Then, color information is used to correct the location of face feature points. Finally, after text mapping, a particular and realistic 3D face model is deformed from a generic model. Using ASM face feature points extraction and correction based on skin color model, the problem of side face information missing is successfully resolved. Depending on only one image and one generic model, the computing cost of memory and time is largely reduced. The 3D face reconstructed can be easily deformed to form different expressions and mouth shapes. Experiments show that this approach is fast and efficient and has an output of a lifelike Chinese talking face.

In this paper an efficient approach of document layout analysis and reading order determination is proposed for a reading robot. Firstly the input document images are preprocessed to remove noises, connect lines and domains, and to reduce the computation time. Secondly a bottom-up, parameter-independent, two-step layout analysis algorithm based on morphology is used, which outlines the geometry of the maximum homogeneous regions and classifies them into texts, tables, and pictures. Finally the reading order is determined, by a top-down recursive hierarchy algorithm derived from XY-cut, using a set of rules depending on layout information. Important parameters are acquired using statistic information of the given images to adapt to different types of documents. The proposed algorithm is applied to a large number of document images and the experimental results show that it makes the reading robot be able to read paper documents of different languages, even with complex layout structure.

In this paper, Hilbert scan based tree representation (HSBT) is presented for image search. Unlike common ways decreasing the number of interest points or reducing the dimensions of features or using searching methods to match interest points, the proposed method builds a tree for each image and gives a new distance measure to calculate the similarity between the query and images in database. In the proposed approach, Hilbert scan for arbitrarily-sized arrays is used to map the interest points from two-dimensional space to one-dimensional space at first. Then, interest points set is divided into several parts by a separation way, and a grouping strategy is given to build a tree for each image. Experimental results show that the proposed approach is space saving. That is because it only stores clustering center and relevant information of each node in the tree. It is also time saving since the similarity calculation is up to the nodes of tree rather than all the descriptors of image. At the same time, the retrieval precision is good, because Hilbert scanning preserves the correlation in two-dimensional image, so nodes of tree are shaped according to the compactness of interest points which can employ the local information as much as possible.

Face finding is a very important initial step towards building up a fully automated face recognition system. Face detection by detecting skin like colors can achieve a high detection rate. In this paper, we presented a novel algorithm for face detection in color images with complex backgrounds. First a parallel structure for skin color detection is proposed to improve the accuracy of detections. The concept of the probability image has been introduced to utilize the color information in the traditional face detection methods specific for gray-scale images. After that, a classifier obtained from Adaboost training is applied to the result of skin detection to reduce the false positives. An experiment has been implemented to verify the improvement of this proposed research. And the proposed approach achieved a better result in this experiment.

One image processing application is to reconstruct the original scene from the low quality images. Considering the idea histogram distribution can reflect good vision effect. So many histogram analyzing based methods have been studied recently. However, some methods require users to set some parameters or condition, and cannot get the optimal results automatically. To overcome those short come, this paper presents an Adaptive Histogram Separation and Mapping (AHSM) method for Backlight image enhancement. First, we separate the histogram by binary tree structure with the proposed Adaptive Histogram Separation Unit (AHSU). And then mapping the Low Dynamic Range (LDR) histogram partition into High Dynamic Range (HDR). By doing this, the excessive or scarcity enhancement can be avoid. The experimental results show that the proposed method can gives better enhancement results, also compared with some histogram analyzing based methods and get better results. © 2010 IEEE.

This paper describes a novel method to match objects in cluttered scenes. This method makes use of Hilbert scanning of feature points in Hough space. We use a 3D Hough transform to obtain a spectrum on which 3D features are concentrated on the sphere. Then, based on the obtained Hough Spectrum, we apply Hilbert scanning on the sphere to match the objects. Using this approach, we can match the object correctly and robustly in both overlapping and noise situation. The characteristic of this method is that it is a global matching method without an estimate of the rotation first and suffering from computational complexity brought by voting/correlation procedure. The experiment results show that the method is more effective compared to existing methods in both matching rate and robustness.

In this work, a novel approach which called 1D local patterns by multi-scans(1DLPMS) for presenting the local features is proposed and its simplifications and extensions to facial image analysis are also considered. First, multi-scans are applied to capture different spatial information on the image with less computation than some traditional ways, such as Local Binary Patterns(LBP). Then, some 1D local patterns are given to encode the local features based on different coding rules. To make the proposed approach computationally simpler and easy to extend, grouped 1D local patterns by multiscans(G1DLPMS) is studied, which divides 1DLPMS into several groups and uses the co-occurrences of these groups. Performance assessment in face recognition under different challenges shows that the proposed approach is superior than traditional ones.

Polar Harmonic Transform (PHT) is termed to represent a set of transforms those kernels are basic waves and harmonic in nature. PHTs consist of Polar Complex Exponential Transform (PCET), Polar Cosine Transform (PCT) and Polar Sine Transform (PST). They are proposed to represent invariant image patterns for two dimensional image retrieval and pattern recognition tasks. They are demonstrated to show superiorities comparing with other methods on describing rotation invariant patterns for images. Kernel computation of PHTs is also simple and has no numerical stability issue. However in order to increase the computation speed, fast computation method is needed especially for real world applications like limited computing environments, large image databases and realtime systems. This paper presents Fast Polar Harmonic Transforms (FPHTs) including Fast Polar Complex Exponential Transform (FPCET), Fast Polar Cosine Transform (FPCT) and Fast Polar Sine Transform (FPST) that are deduced based on mathematical properties of trigonometric functions. The proposed FPHTs are averagely over 6 similar to 8 times faster than PHTs that significantly boost computation process. The experimental results on both synthetic and real data are given to illustrate the effectiveness of the proposed fast transforms.

In-vehicle cameras are widely used to obtain visual information around a car. In good weather conditions, high visibility images can be obtained by using the cameras. However they do not work usefully in bad conditions such as rain. In this paper, we propose a method for removing raindrops on a windshield and repairing their regions in an image sequence captured by a vehicle mounted monocular camera. We also discuss image restoration for car-mounted camera images in bad weather conditions.

Human face detection plays an important role in many application areas such as video surveillance, human computer interface, face recognition, face search and face image database management etc. In human face detection applications, face region usually form an inconsequential part of images. Consequently, preliminary segmentation of images into regions that contain "non-face" objects and regions that may contain "face" candidates can greatly accelerate the process of human face detection. Color information based methods take a great attention, because colors have obviously character and robust visual cue for detection. This paper proposed a new method based on RGB color centroids segmentation (CCS) for face detection. This paper include two parts, first part is color image thresholding based on CCS and the second part is face detection based on region growing and facial features structure character combined method. The experimental results show the ideal thresholding result and better than the result of other color space analysis based thresholding methods. Proposed method can conquer the influence of different background conditions, position, scale instance and orientation in images from several photo collections and database; the effect is also better than existing skin color segmentation based methods.

Super resolution is a technique of enhancing image resolution by combining information from multiple images. It is widely applied in fields like camera surveillance, satellite imaging, pattern recognition, etc. One challenging problem of super resolution is its high demand on image registration accuracy. This paper introduces a high accuracy registration approach for the purpose of super resolution. It is invariant to translation, scaling, rotation, and noise, and can be used to automatically obtain the Maximize a Likelihood Estimation (MLE) of image homography (registration result) using information only contained within the images themselves. An effective Genetic Algorithm based approach is used to filter out all the mismatches. Comparison with RANSAC and Keren's method will be given to prove the effectiveness of the proposed method.

This paper presents a novel algorithm for face recognition. Local Quaternion Patters (LQP) is proposed for presenting the feature parts in the face. To keep the spatial feature of the face, an asymmetric similarity measure Weighted Spatially constrained Earth Mover's Distance (WSEMD) is studied for classification. In this step, the source image is partitioned into non overlapping local patches while the destination image is represented as a set of overlapping local patches at different positions and Gaussian Kernel is used. Finally, local and global weighting is applied to get a more accurate classifier. To evaluate the proposed method and its performance, three well-known and challenge face databases - ORL, Yale and FERET are used in our study. The experimental results show that the proposed method has higher accuracy than some other classic methods.

Signature verification is a challenging task, because only a small set of genuine samples call be acquired and usually no forgeries are available in real application. In this paper, we propose a novel approach based oil Hilbert scanning patterns and Gaussian mixture models for automatic on-line signature verification. Our system is composed of a similarity measure based oil Hilbert scanning patterns and a simplified Gaussian mixture model for decision-level evaluation. To be practical, we introduce specific simplification strategies for model building and training. The system is compared to other state-of-the-art systems based oil the results of the First International Signature Verification Competition (SVC 2004). Experiments are conducted to verify, the effectiveness of our system.

Real world scenes contain a large range of light intensities which range from dim starlight to bright; sunlight. A common task of tone mapping algorithms is to reproduce high dynamic range(HDR) images on low dynamic range(LDR) display devices such as printers and monitors. In this paper, a new tone mapping algorithm is proposed for the display of HDR images. Inspired by the adaptive process of the human visual system, the proposed algorithm utilized the center-surround Retinex processing. The novelty of our method is that the local details are enhanced according to a non-linear adaptive spatial filter (Gaussian filter), whose shape is adapted to high-contrast edges of the image. The proposed method uses an adaptive surround instead of the traditional pre-defined circular. Therefore, the algorithm can preserve visibility and contrast impression of high dynamic range scenes in the common display devices. The proposed method is tested on a variety of HDR, images, and we also compare it to previous work. The results show good performance of our method in terms of visual quality.

Super resolution is a technique of enhancing image resolution by combining information from multiple images. It is widely applied in fields like camera surveillance, satellite imaging, pattern recognition, etc. One challenging problem of super resolution is its high demand on image registration accuracy. This paper introduces a high accuracy registration approach for the purpose of super resolution. It is invariant to translation, scaling, rotation, and noise, and can be used to automatically obtain the Maximize a Likelihood Estimation (MLE) of image homography (registration result) using information only contained within the images themselves. An effective Genetic Algorithm based approach is used to filter out all the mismatches. Comparison with RANSAC and Keren's method will be given to prove the effectiveness of the proposed method. ©2009 IEEE.

Signature verification is a challenging task, because only a small set of genuine samples can be acquired and usually no forgeries are available in real application. In this paper, we propose a new function-based method with Hilbert Scanning patterns for automatic on-line signature verification. The proposed method is compared to other state-of-the-art methods based on the results of the First International Signature Verification Competition (SVC 2004). Experiments are conducted to verify the effectiveness of our method. ©2009 IEEE.

In this study, we try to read printed characters on the book. This paper describes the method to segment a sentence to characters and search a pronunciation of the character by the dictionary.

In this work, two novel local feature patterns-Modified Local Binary patterns (MLBP) and local Ternary patterns (LIP), are proposed for extract features in the facial image, which use some distinct rule to code the values in a label, respectively. These patterns are more invariant to illuminance and face expression compared to traditional one. After getting the local feature patterns, in order to take alignment of face into account, a novel matching method called Histogram Spatially constrained Earth Mover's Distance(HSEMD) is proposed. In this step, the source image is partitioned into non-overlapping local regions while the destination image is represented as a set of overlapping local regions at different positions. Meanwhile, multi-scale cascade mechanism is studied for extracting more feature patterns and obtaining global information of the face.The performance of the proposed method is assessed in the face recognition problem under different challenges. The experimental results show that the proposed method has higher accuracy than some other classic methods.

Recently, video lossless compression has been developed for applying it to digital cinema, video archiving of contents, etc. Video lossless compression is important in image processing problem since a large image requires a large amount of storage space. The purpose of this paper is to enhance the predictor used for the lossless compression of video. In this paper, we propose the 3-dimensional predictor for the effective prediction. In addition, the three-dimensional spatio-temporal gradient is adopted to improve the conventional image compression methods such as GAP, MED which are two-dimensional predictions based on horizontal and vertical gradients. The spatio-temporal gradient is a spatial data resulted from the projection of triangular prism composed of the neighborhood pixels. From the experimental results compared with the previous prediction methods, we confirmed that the prediction using proposed method is more efficient.

NIR is an open source cloud computing 'enabled content based Image retrieval system With the development and popularization of cloud computing, more and more researchers from different research areas do research with the help of cloud computing Nowadays content based image retrieval as one of the challenging and emerging technologies is high computation task because of the algorithm computation complexity and big amount of data As based on cloud computing infrastructure, NIR is easy to extent and flexible for deployment As an open source project, NIR can be improved on demand and integrated to other existing systems This paper presents our ideas, findings, design and the system from our work of NIR

This paper presents a 3D face recognition algorithm using fast landmark detection and non-rigid iterative Closest Point (ICP) algorithm. The proposed approach can estimate the facial feature region using the anthropometric face model after pose correction, and accurately detect 9 facial landmarks (nose tip, sellion, inner and outer eye corners, nostrils and mouth center). An extension of ICP algorithm has also been proposed to matching the non-rigid 3D face shapes. Experimental results demonstrate that compared to the existing methods, the proposed approach can efficiently detect human facial landmarks and satisfactorily deal with the 3D face matching problem.

This paper addressed the problem of multi-frame image super resolution and its implementation to fingerprint image.,The use of computers in fingerprints recognition is highly desirable in many applications where security is an important concern. However, as the performance of fingerprint recognition algorithm relies heavily on the quality of the input fingerprint images, it is essential to incorporate a fingerprint enhancement algorithm in the minutiae extraction module. Typical approaches use Gabor filter to raise the contrast between fingerprint ridges and valleys. But few of them try to increase the very basic resolution of fingerprint image. Therefore, we proposed a computer vision solution to this problem which involves the technique of super resolution. This approach can be incorporated into any traditional fingerprint enhancement algorithm as a preprocessing step, rendering a better overall enhancement result.

An algorithm for automatic image-map alignment problem using a new similarity measure named Edge-Based Code Mutual Information (EBCMI) and Hilbert scan is presented in this study. Because image and map are very different in their representations, the normal Mutual Information (MI) using the intensity in traditional alignment method may result in misalignment. To solve the problem, codes which are robust to the differences between the image-map pairs are constructed and Mutual Information of the codes is computed as the similarity measure for the alignment. We convert the 3-D transformation search space in alignment to a 1-D search space sequence by using 3-D Hilbert Scan. A new search strategy is also proposed on the 1-D search space sequence. The experimental results show that the proposed EBCMI outperformed the normal MI and some other similarity measures and the proposed search strategy gives flexibility between efficiency and accuracy for automatic imagemap alignment task.

In this paper, we propose a method for extracting raindrops based on analysis on motion vectors. The proposed method is applied to extract raindrops on a water repellent windshield from in-vehicle camera images. As a result of experiments, apporoximately 40 percent of raindrops in an image are extracted by our method.

In predictive coding, prediction methods that aim to more efficient prediction by switching some predictors based on local index have been proposed in the past. Difference between adjacent pixels, i.e. gradients, is popularly used in such methods. However, relationship between the gradients and predictors is discussed only qualitatively. In this paper, we propose a prediction method using difference to average prediction value as local index. In the experiments, the proposed prediction scheme shows its superiority to LOCO-I and CALIC in terms of entropy of prediction error and code length.

Point Pattern Matching (PPM) is an essential problem in many image analysis and computer vision tasks. This paper presents a two-stage algorithm for PPM problem using ellipse fitting and dual Hilbert scans. In the first matching stage, transformation parameters, are coarsely estimated by using four node points of ellipses which are fitted by Weighted Least Square Fitting (WLSF). Then, Hilbert scans are used in two aspects of the second matching stage: it is applied to the similarity measure and it is also used for search space reduction. The similarity measure named Hilbert Scanning Distance (HSD) can be computed fast by converting the 2-D coordinates of 2-D points into 1-D space information using Hilbert scan. On the other hand, the N-D search space can be converted to a 1-D search space sequence by N-D Hilbert Scan and art efficient search strategy is proposed on the 1-D search space sequence. In the experiments. we use both simulated point set data and real fingerprint images, to evaluate the performance of our algorithm and our algorithm gives satisfying results both in accuracy and efficiency.

Signature-based personal identification systems are used and accepted widely due to its distinctness and stability. In this paper, we propose a new function-based method with Hilbert-Scanning patterns for signature matching. We evaluate the performance by using the online signature database Signature Verification Competition (SVC) 2004.

In this paper, we propose video lossless compression based on 3-dimensional prediction using spatio-temporal gradients. The proposed method predicted the target pixel using the LGM(Local Gradient Magnitude), which is the gradient between the neighborhood pixel and the current pixel in the spatio-temporal. From the experimental results, we demonstrated that the spatio-temporal gradients prediction is more efficient.

The N-dimensional (N-D) Hilbert curve is a one-to-one mapping between N-D space and one-dimensional (I-D) space. It is studied actively in the area of digital image processing as a scan technique (Hilbert scan) because of its property of preserving the spatial relationship of the N-D patterns. Currently there exist several Hilbert scan algorithms. However. these algorithms have two strict restrictions in implementation. First, recursive functions are used to generate a Hilbert curve, which makes the algorithms complex and computationally expensive. Second, all the sides of the scanned region must have the same size and the length must be a power of two, which limits the application of the Hilbert scan greatly. Thus in order to remove these constraints and improve the Hilbert scan for general application, a nonrecursive N-D Pseudo-Hilbert scan algorithm based on two look-up tables is proposed in this paper. The merit of the proposed algorithm is that implementation is much easier than the original one while preserving the original characteristics. The experimental results indicate that the Pseudo-Hilbert scan can preserve point neighborhoods as much as possible and take advantage of the high correlation between neighboring lattice points, and it also shows the competitive performance of the Pseudo-Hilbert scan in comparison with other common scan techniques. We believe that this novel scan technique undoubtedly leads to many new applications in those areas can benefit from reducing the dimensionality of the problem.

A new fingerprint matching algorithm using dual Hilbert scans is presented in this study. We treat the fingerprint matching as point pattern matching problem and Hilbert scans are used in two aspects of the matching problem: one is applied to the similarity measure and the other is used in search space reduction. The similarity measure named Hilbert Scanning Distance (HSD) can be computed fast by converting the 2-D coordinates of 2-D images into 1-D space information using Hilbert scan. On the other hand, the 3-D search space can be converted to a 1-D search space sequence. The proposed method has been tested on FVC2002 database. The experimental results show that our method can implement fingerprint matching robustly and efficiently. The performance evaluation EER (Equal-Error Rate) generally used is very low by our algorithm.

This study presents a new algorithm for automatic image-map alignment problem using a new similarity measure named Edge-Based Code Mutual Information (EBCMI) and 3-D Hilbert scan. In general, each image-map pair can be viewed as two special multimodal images, however, are very different in their representations such as the intensity. Therefore, the normal Mutual Information (MI) using the intensity in traditional alignment method may result in misalignment. To solve the problem, codes based on the edges of the image-map pairs are constructed and Mutual Information of the codes is computed as the similarity measure for the alignment in our method. Since Edge-Based Code (EBC) is robust to the differences between the image-map pairs in their representations, EBCMI also can overcome the differences. On the other hand, the 3-D search space in alignment can be converted to a 1-D search space sequence by 3-D Hilbert Scan and a new search strategy is proposed on the 1-D search space sequence. The experimental results show that the proposed EBCMI performed better than the normal MI and some other similarity measures and the proposed search strategy gives flexibility between efficiency and accuracy for automatic image-map alignment task. © 2008, The Institute of Image Electronics Engineers of Japan. All rights reserved.

Image enhancement plays an important role in many machine vision applications on images captured in low contrast and low illumination conditions. In this study, we propose a new method for image enhancement based on analysis on embedded surfaces of images. The proposed method gives an insight into the relationship between the image intensity and image enhancement. In our method, scaled surface area and the surface volume are proposed and used to reconstruct the image iteratively for contrast enhancement, and the illumination of the reconstructed image can also be adjusted simultaneously. On the other hand, the most common methods for measuring the quality of enhanced images areMean Square Error (MSE) or Peak Signal-to-Noise-Ratio (PSNR) in conventional works. The two measures have been recognized as inadequate ones because they do not evaluate the result in the way that the human vision system does. This paper also presents a new framework for evaluating image enhancement using both objective and subjective measures. This framework can also be used for other image quality evaluations such as denoising evaluation. We compare our enhancement method with some well-known enhancement algorithms, including wavelet and curvelet methods, using the new evaluation framework. The results show that our method can give better performance in most objective and subjective criteria than the conventional methods. Copyright © 2008 The Institute of Electronics, Information and Communication Engineers.

Road sign detection is one of the major concerned topics in the field of driving safety and intelligent vehicle. In this paper, a novel model based on Color Barycenters Hexagon (CBH) is proposed and used to defect road sign usefully. In CBH model, full color images are calculated the color barycenters and get the barycenters region, then automatic select the idea threshold curves to separate the Region of Interest (ROI) of barycenters aiming to detect the road sign. Because of the practically images have many noise, and the existing color space cannot separate the ROI ideally. The proposed CBH model can thresholding the principal color of ROI and have high robust. With suitably thresholding and operations, road sign on various scene images can be detected.

Image enhancement is important for images captured in low contrast and low illumination conditions. In this study, we propose a new iterative algorithm for image enhancement based on analysis on embedded surfaces of images. In our method, scaled surface area and the surface volume are proposed and used to reconstruct the image iteratively for contrast enhancement, and the illumination of the reconstructed image can also be adjusted simultaneously. On the other hand, the most common methods for measuring the quality of enhanced images Mean Square Error (MSE) or Peak Signal-to-Noise-Ratio (PSNR) have been recognized as inadequate measures because they do not evaluate the result in the way that the human vision system does. This paper also presents a new framework for evaluating image enhancement using both objective and subjective measures. This framework can also be used for other image quality evaluations such as denoising evaluation. We compare our enhancement method with some well-known enhancement algorithms, including wavelet and curvelet methods, using the new evaluation framework. The results show that our method gives better performance in most objective and subjective criteria than the conventional methods.

In this paper we present a new tone mapping algorithm for the display of high dynamic range images, inspired by adaptive process of the human visual system. The proposed algorithm is based on the center-surround Retinex processing. In our method, the local details are enhanced according to a non-linear adaptive spatial filter (Gaussian filter), whose shape (filter variance) is adapted to high-contrast edges of the image. Thus our method does not generate halo artifacts meanwhile preserves visibility and contrast impression of high dynamic range scenes in the common display devices. The proposed method is tested on a variety of HDR images and the results show the good performance of our method in terms of visual quality.

The Hilbert curve is one of space-filling curves published by G. Peano. There are several applications using this curve, such as image processing, computer graphics, etc. In this paper, we concentrate on a tone mapping technique for high dynamic range images using the Pseudo-Hilbert curve. Based on the neighbourhood property of the Pseudo-Hilbert scan, a fast and flexible tone reproduction method is proposed. The proposed new technique preserves visibility and contrast impression of high dynamic range scenes in low dynamic range reproduction devices. From the experimental results, we have confirmed that the proposed method produces good results on a variety of high dynamic range images.

Image contrast enhancement plays an important role in many machine vision applications. In this study, we propose a new method for edge enhancement based on analysis on embedded surfaces of images. The proposed method gives an insight into the relationship between the image intensities (also the gradients) and image contrast. In our method, scaled surface area and the surface volume are used to reconstruct the image for edge enhancement, and then the contrast of the reconstructed image is adjusted by a 'strengthen-weaken' process. Although, current method for edge enhancement such as curvelet transform can enhance the edge part, it does not provide good tonal rendition or color constancy sometimes. The experimental results show that our method can give good performance not only in edge enhancement, but also in tonal rendition and color constancy.

Recently, video lossless compression is developed for purpose of applications to digital cinema, video archiving of contents, etc. In this paper, we present video lossless compression based on three-dimensional prediction using spatio-temporal gradients. By extending previous image compression methods about two-dimensional prediction based on horizontal and vertical gradients, such as GAP, MED, we utilize three-dimensional spatio-temporal gradients to prediction. From the experimental results compared with the previous prediction techniques, we confirmed that the spatio-temporal prediction is more efficient.

Sewer environment is composed of cylindrical pipes, in which only a few landmarks such as manholes, inlets and pipe joints are available for localization. This paper presents a method for navigation of an autonomous sewer inspection robot in a sewer pipe system based on detection of landmarks. In this method, location of an autonomous sewer inspection robot in the sewer pipe system is estimated from stereo camera images. The laser scanner data are also used to ensure accurate localization of the landmarks and reduce the error in distance estimation by image processing. The method is implemented and evaluated in a sewer pipe test field using a prototype robot, demonstrating its effectiveness.

The 2-dimensional (2-D) Hilbert curve is a one-to-one mapping between 2-D space and one-dimensional (1-D) space. It is studied actively in the area of digital image processing as a scan technique (Hilbert scan) because of its property of preserving the spacial relationship of the 2-D patterns. Currently there exist several Hilbert scan algorithms. However, these algorithms have two strict restrictions in implementation. First, recursive functions are used to generate a Hilbert curve, which makes the algorithms complex and computationally expensive. Second, both sides of the scanned rectangle must have same size and each size must be a power of two, which limits the application of the Hilbert scan greatly. In this paper, a Pseudo-Hilbert scan algorithm based on two look-up tables is proposed. The proposed method improves the Hilbert scan to be suitable for real-time processing and general application. The simulation indicates that the Pseudo-Hilbert scan can preserve point neighborhoods as much as possible and take advantage of the high correlation between neighboring lattice points. It also shows competitive performance of the Pseudo-Hilbert scan in comparison with other scan techniques.

In this study, we propose a simple, yet general and powerful framework for constructing accurate affine invariant regions. In our framework, a method for extracting reliable seed points is first proposed. Then, regions which are invariant to most common affine transformations can be extracted from seed points by two new methods the Path Growing (PG) or the Thresholding Seeded Growing Region (TSGR). After that, an improved ellipse fitting method based on the Direct Least Square Fitting (DLSF) is used to fit the irregularly-shaped contours from the PG or the TSGR to obtain ellipse regions as the final invariant regions. In the experiments, our framework is first evaluated by the criterions of Mikolajczyk's evaluation framework [1], and then by near-duplicate detection problem [2]. Our framework shows its superiorities to the other detectors for different transformed images under Mikolajczyk's evaluation framework and the one with TSGR also gives satisfying results in the application to near-duplicate detection problem. Copyright © 2007 The Institute of Electronics, Information and Communication Engineers.

The 3-dimensional (3-D) Hilbert scan is a one-to-one mapping between 3-D data and 1-D data along the 3-D Hilbert curve. It has been applied widely in image processing, such as image compression, object recognition, and image clustering, etc. Now, although there exist some 3-D Hilbert scanning algorithms, they usually have strict limitation on the scanned region. This makes Hilbert scan difficult to be applied in practice. So an effective scanning algorithm for arbitrarily-sized cuboid region is significant to improve the correlative digital image processing technology. In this paper, we proposed a novel Pseudo-Hilbert scanning algorithm based on the look-up tables method for arbitrarily-sized cuboid region. Although the proposed algorithm is designed for 3-D space scanning, it can be also applied in an arbitrary-sized rectangle. The algorithm does not only remove the strict constrains but also reserve the good property of the Hilbert curve preserving point neighborhoods as much as possible. The good performance of the algorithm is demonstrated by the simulation results. © 2006 IEEE.

In this study, we propose a simple, yet general and powerful framework for constructing accurate affine invariant regions and use it for near-duplicate detection problem. In our framework, a method for extracting reliable seed points is first proposed. Then, regions which are invariant to most common affine transformations are extracted from seed points by a new method named the Thresholding Seeded Growing Region (TSGR). After that, an improved ellipse fitting method based on the Direct Least Square Fitting (DLSF) is used to fit the irregularly-shaped contours of TSGRs to obtain ellipse regions as the final invariant regions. At last, SIFT-PCA descriptors are computed on the obtained regions. In the experiment, our framework is evaluated by retrieving near-duplicate in an image database containing 1000 images. It gives a satisfying result of 96.8% precision at 100% recall.

The N-dimensional (N-D) Hilbert curve is a oneto-one mapping between N-D space and one-dimensional (I-D) space. It is studied actively in the area of digital image processing as a scan technique (Hilbert scan) because of its property of preserving the spacial relationship of the N-D patterns. Currently there exist several Hilbert scan algorithms. However, these algorithms have two strict restrictions in implementation. First, recursive functions are used to generate a Hilbert curve, which makes the algorithms complex and computation ally expensive. Second, all the sides of the scanned region must have same size and each size must be a power of two, which limits the application of the Hilbert scan greatly. In this paper, a nonrecursive N-D Pseudo-Hilbert scan algorithm based on two look-up tables is proposed. The merit of the algorithm is that the computation is fast and the implementation is much easier than the original one. The simulation indicates that the Pseudo-Hilbert scan can preserve point neighborhoods as much as possible and take advantage of the high correlation between neighboring lattice points. It also shows competitive performance of the Pseudo-Hilbert scan in comparison with other common scan techniques.

In this paper, we propose a descriptor which is invariant to general deformations (only intensity locations change but not their value) by using Hilbert scanning. In our method, an image is converted to a 1-D sequence through Hilbert scanning at first. Then, we embed this sequence as a 1-D curve in the 2-D space. Because Hilbert scanning preserves the coherence in a 2-D image, it is easily to understand that the area under the curve is invariant to intensity location changes, naturally. Hence, we use some areas for an interest point as a deformation invariant descriptor. The experimental results show that the dimension of our descriptor is low and it is superior to other approaches on interest point matching in deformation images.

The three-dimensional Hilbert curve is a one-to-one mapping between three-dimensional (3-D) space and one-dimensional (1-D) space. Due to the advantage of preserving high correlation of 3-D points, it receives much attention in many areas. Especially as a scan technique (Hilbert scan), the Hilbert curve is studied actively in image processing. Although there exist several Hilbert scan algorithms now, they have strict implementation conditions. First, they use recursive functions to generate scanning curves, which makes the algorithms complex and difficult to implement in real-time systems. Second, the scanned region must be a square and the length must be the power of two, which limits the application of the Hilbert scan greatly, In this paper, an effective 3-D Pseudo-Hilbert scan algorithm for an arbitrarily-sized cuboid is proposed. The proposed algorithm improves the Hilbert scan for general application. Moreover, it greatly reduces the computational complexity and saves storage memory by using two simple look-up tables instead of recursive functions. Therefore the algorithm is suitable for real-time processing. The experimental results show that the Pseudo-Hilbert scan preserves the most structures of the Hilbert scan. Although the proposed algorithm is 3-D case, it is also feasible in 2-D space. We believe this novel scan technique undoubtedly leads to many new applications in those areas which can benefit from reducing the dimensionality of the problem.

Natural, continuous tone images have a very important property of high correlation of adjacent pixels. Images which we wish to compress are usually non-stationary and can be reasonably modeled as smooth and textured areas separated by edges. This property has been successfully exploited in LOCO-I and CALIC by applying gradient based predictive coding as a major de-correlation tool. However, they only examine the horizontal and vertical gradients, and assume the local edge can only occur in these two directions. Their over-simplified assumptions hurt the robustness of the, prediction in higher complex areas. In this paper, we propose an accurate gradient selective prediction (AGSP) algorithm which is designed to perform robustly around any type of image texture. Our method measures local texture information by comparison and selection of normalized scalar representation of the gradients in four directions. An adaptive predictor is formed based on the local gradient information and immediate causal pixels. Local texture properties are also exploited in the context modeling of the prediction error. The results we obtained on a test set of several standard images are encouraging. On the average, our method achieves a compression ratio significantly better than CALIC without noticeably increasing of computational complexity.

In this report, we present a novel similarity measure using Hilbert curve for point pattern matching. In our method, the similarity measure is computed in one-dimensional (1-D) sequence in stead of in two-dimensional (2-D) space by using Hilbert curve. The experimental results show that our measure is fast and robust to noise than conventional similarity measures.

In this paper, we present three novel lossless compression approaches for gray-scale continuous tone natural image. Our methods enhance the median edge detector (MED), which is the core part of JPED-LS algorithm, by reducing the entropy of the prediction error via adaptive regression. These modified predictors improve the prediction accuracy by reducing the negative effect due to MED's oversimplified edge orientation detection. The experimental results show that our approaches achieve evidently better performance than MED with only neglectable increasing of computational complexity and without introduce extra pixels into the causal template.

To find the best transformation between a "model" point set and an "image" point set is the main purpose of point pattern matching. The similarity measure plays a pivotal role and is used to determine the degree of resemblance between two objects. Although some well-known Hausdorff distance measures work well for this task, they are very computationally expensive and suffer from the noise points. In this paper, we propose a novel similarity measure using the Hilbert curve named Hilbert scanning distance (HSD) to resolve the problems. This method computes the distance measure in the one-dimensional (1-D) sequence instead of in the two-dimensional (2-D) space, which greatly reduces the computational complexity. By applying a threshold elimination function, large distance values caused by noise and position errors (e.g. those that occur with feature or edge extraction) are removed. The proposed algorithm has been applied to the task of matching edge maps with noise. The experimental results show that HSD can provide sufficient information for image matching within low computational complexity. We believe this sets a new direction for the research of point pattern recognition.

The 2-dimensional Hilbert scan (HS) is a one-to-one mapping between 2-dimensional (2-D) space and one-dimensional (1-D) space along the 2-D Hilbert curve. Because Hilbert curve can preserve the spatial relationships of the patterns effectively, 2-D HS has been studied in digital image processing actively, such as compressing image data, pattern recognition, clustering an image, etc. However, the existing HS algorithms have some strict restrictions when they are implemented. For example, the most algorithms use recursive function to generate the Hilbert curve, which makes the algorithms complex and takes time to compute the one-to-one correspondence. And some even request the sides of the scanned rectangle region must be a power of two, that limits the application scope of HS greatly. Thus, in order to improve HS to be proper to real-time processing and general application, we proposed a Pseudo-Hilbert scan (PHS) based on the look-up table method for arbitrarily-sized arrays in this paper. Experimental results for both HS and. PHS indicate that the proposed generalized Hilbert scan algorithm also reserves the good property of HS that the curve preserves point neighborhoods as much as possible, and gives competitive performance in comparison with Raster scan.

The purpose of image-map registration is to revise the digital map included in Geographic Information System (GIS) with aerial image. The traditional method for this task requires the manual selection of tie points in both image and map. In this study, we propose a distance measure named Voting Weighted Modified Hausdorff Distance (VWMHD) for this task. In order to overcome the differences in representations between image and map in urban area, after several times of edge extraction through multiscale space, we give weights to each edge point in the initial scale based on its voting times and then compute the VWMHD for registration. The experimental results implicate that our VWMHD can provide sufficient information for automatic image-map registration and is robust to noises.

The purpose of image-map registration is to revise the digital map included in Geographic Information System (GIS) with aerial image. The traditional method for this task requires the manual selection of features such as tie points in both the image and the map. Determining how to automatically extract common features between image and map is a difficult problem. In this study, we propose a novel framework of extracting the common feature named Diffusion Geodesic Path (DGP) between image and map, and use it for image-map registration. In order to overcome the differences in presentations between image and map, we first construct two new edge images from the image and the map using the diffusion process in physics to diffuse the differences between them. Then, we extract several paths between feature points (corner points) on these new edge images using the geodesic. Finally, a part of the extracted paths are automatically selected and are used for registration. The experimental results implicate that our DGP is a common and robust feature between image and map, and it can provide sufficient information for automatic image-map registration. The whole framework works well on automatic image-map registration.

In order to maximize the robustness of fingerprint identification and to minimize the identification time, it is better to align the fingerprint first by referring the singular points of fingerprint. In this paper, we proposed a new algorithm to localize the singular points of fingerprint.
The proposed approach is less computational complexity and higher precision. This paper shows how, from the x-gradient component of orientation filed and its sine component, very accurate extraction of singular points can be obtained. This localization can be used for accurate registration of two fingerprints.

In this paper, we present three novel lossless compression approaches for gray-scale continuous tone natural image. Our methods enhance the median edge detector (MED), which is the core part of JPED-LS algorithm, by reducing the entropy of the prediction error via adaptive regression. These modified predictors improve the prediction accuracy by reducing the negative effect due to MED's oversimplified edge orientation detection. The experimental results show that our approaches achieve evidently better performance than MED with only neglectable increasing of computational complexity and without introduce extra pixels into the causal template.

In this paper, we propose a descriptor which is invariant to general deformations (only intensity locations change but not their value) by using Hilbert scanning. In our method, an image is converted to a 1-D sequence through Hilbert scanning at first. Then, we embed this sequence as a 1-D curve in the 2-D space. Because Hilbert scanning preserves the coherence in a 2-D image, it is easily to understand that the area under the curve is invariant to intensity location changes, naturally. Hence, we use some areas for an interest point as a deformation invariant descriptor. This descriptor can be computed in the 2-D space efficiently than other approaches where an image is embedded in the 3-D space or the dimensions of descriptors are very large. The experimental results show that our descriptor is low-complexity and superior to other approaches on interest point matching in deformation images.

A fast and accurate similarity named Hilbert Scanning Distance(HSD) [9] has recently been presented for point matching. In this study, we improved an efficient algorithm of search strategy for HSD in the large search space. This search strategy is associated with two ideas: a relaxation greedy search, and an accelerating process using Monte Carlo sampling. The experimental results implicate that this improved algorithm is robust and efficient for point matching using HSD. It also makes a tradeoff between accuracy and speed under different requirements.

In order to maximize the robustness of fingerprint identification and to minimize the identification time, it is better to align the fingerprint first by referring the singular points of fingerprint. In this paper, we proposed a new algorithm to localize the singular points of fingerprint.
The proposed approach is less computational complexity and higher precision. This paper shows how, from the x-gradient component of orientation filed and its sine component, very accurate extraction of singular points can be obtained. This localization can be used for accurate registration of two fingerprints.

The purpose of image-map registration is to revise the digital map included in Geographic Information System (GIS) with aerial image. The traditional method for this task requires the manual selection of features such as tie points in both the image and the map. Determining how to automatically extract common features between image and map is a difficult problem. In this study, we propose a novel framework of extracting the common feature named Diffusion Geodesic Path (DGP) between image and map, and use it for image-map registration. In order to overcome the differences in presentations between image and map, we first construct two new edge images from the image and the map using the diffusion process in physics to diffuse the differences between them. Then, we extract several paths between feature points (corner points) on these new edge images using the geodesic. Finally, a part of the extracted paths are automatically selected and are used for registration. The experimental results implicate that our DGP is a common and robust feature between image and map, and it can provide sufficient information for automatic image-map registration. The whole framework works well on automatic image-map registration.

This paper presents a new padding technique for arbitrarily shaped coding. Padding background pixels using pixels in shapes, square blocks are generated; therefore, traditional orthogonal transforms work availably. A padding technique called low-pass extrapolation (LPE) is indeed employed in MPEG-4. Shen and Liou demonstrate a sophisticated padding for 1D-DCT, which guarantees that as many high frequency DCT coefficients as background pixels become zero. In order to apply their method to 2D-DCT, they also show two solutions. However, the solutions do not accomplish the equivalence between the numbers of the high frequency coefficients and the background pixels because the proposals are merely based on the 1D-DCT. In this paper, we enhance their padding methods to the orthogonal transform used in H.264/AVC, namely integer DCT. In the proposed method, the number of the DCT coefficients to be coded is equivalent to the number of the pixels lie in the shape and is appear in the zigzag order.

Finding significant change in high resolution sensed image is an important task in maintaining GIS database. A class of these algorithms detects changed regions by means of edge comparisons. After extraction of feature points from a sensed image and a reference image, the feature points matching is a pivotal key in change detection. In general, given two point sets, find the minimum or maximal value of some measuring distances under the (affine) transformation. Because of the measurement errors and some outlying points, it is important that the measuring distances should be robust. Recently, a well known robust measuring distance called (partial) Hausdorff distance is widely used in feature points matching. It is more efficient than other conventional methods and has been applied in many fields. Although it is a reliable similarity measure, it is also a computational task. In this paper, we present a new algorithm using Hubert curve in order to resolve the computational complexity problem. This distance can be computed in the 1-D space rather than in the 2-D space that can greatly reduce the computational complexity. Our algorithm shows good performances for this task.

Huge amount of high resolution satellite images for remote sensing are available in a couple of decades. Image compression method is required or highly desired in this task. Image coding standards such as JPEG and MPEG are widely used for compressing these images. However, JPEG (or MPEG) compressed data stream can not be processed without decoding. In this paper, we concentrate on image compression and retrieval without decoding by using Hilbert curve. Hubert curve is one of space filling curves which pass through all points in a space. The merits of our compression method is that (1) the compression ratio is almost equivalent to JPEG2000, (2) the computation is quite simple, (3) the required memory is rather small. This paper presents that our compression algorithm can be utilized for image retrieval from image database without decoding the compressed data stream. The block color histogram generated from the compressed data stream is used to compute the similarity measure between two images. The experimental results show that our approaches achieve better performance than conventional methods.

The purpose of image-map registration is to revise the digital map included in Geographic Information System (GIS) with Remote Sensing Images (RSI). Because edge points extracted from RSI are very different from those extracted from digital maps, the error of registration different edge points propagates to the final registration results. In this paper, we provide an automatic algorithm for the image-map registration problem. A novel concept of modified Partial Hausdorff distance (MPHD) is proposed its :I distance measure. This measure is quite tolerant of the differences between map and image. In order to decrease the computational complexity, we also use Hilbert curve to select the initial candidates of edge points in our algorithm. The number of initial candidates can be adjusted to meet the required search accuracy and/or speed. The experimental results implicate that MPHD of our algorithm could provide sufficient information for image-map registration.

Stereo matching is an essential issue in computer vision. Recently, many stereo matching algorithms based on segmentation, graph cuts and so on have been proposed. Because the disparities change continuously in sewer environment, these methods are not applicable to sewer systems and are computationally expensive. In this paper, we propose a new cooperative stereo matching algorithm that is suitable for this task. In our algorithm, a reference image is divided into a feature pixel group (edges, cracks, etc.) and a non-feature pixel group (walls, etc.). Then, a proposed matching measure named Linear Computation is implemented in the feature pixel group. Also, the conventional measure of the Sum of Squared Differences (SSD) is implemented in the non-feature pixel group. In order to improve the accuracy of the conventional SSD measure, we also impose constraints such as an evidence constraint and a neighboring similarity constraint on the SSD measure. The proposed cooperative algorithm achieves better results than other conventional algorithms in the sewer environment.

Matching objects is a fundamental problem for any object detection system. Feature-based methods in matching objects such as vehicles often encounter the problem of correspondences between features of two related patterns. The features may be points, fines, curves and regions. Point pattern matching (PPM) is a primary and essential approach for establishing a correspondence within two related patterns. Although some well-known Hausdorff distance measures work well for this task, they are very computational expensive and suffer from the noise of images. In this paper, we propose a novel similarity measure using Hilbert curve named Hilbert scanning distance (HSD) to resolve the problems. This method computes the distance measure in one-dimensional (I-D) sequence in stead of in two-dimensional (2-D) image space, which greatly reduce the computational complexity. By applying a threshold elimination function, extreme distances caused by noise and position errors (e.g. those that occur with feature or edge extraction) are removed. The experimental results show that HSD can provide sufficient information for matching vehicles within low computational complexity. We believe this point out a new direction for the research of PPM.

In this paper we propose a method for an autonomous sewer robots to navigate through a sewer pipe system based on stereo camera information. In this method, a local features such as manholes and pipe joints are extracting as a feature pixels in the Region of Interest (ROI) of left image. Then, an accurate and fast stereo matching measure named Linear Computation is implemented in this ROI image to compute the distance between the robots and local features. Finally, the distance data can be used for navigation map in sewer pipe system. The experimental results show that our method can provide sufficient information for an autonomous sewer robots navigation.

In 1890, G. Peano found a curve which passes through all points in a space and proved its existance. So far such a countinuous curve which maps a unit interval into a unit hypercube is called a space-filling curve (SFC) or a Peano curve. There are several applications in the area of image processing, computer graphics, database retrieval, etc. Among the SFC's, the most applicable curve is a Hilbert curve. This paper describes the definition and some examples of the SFC, and then overviews image processing applications of the SFC, especially the Hilbert curve.

Predictive coding is generally employed in lossless coding algorithms for digitized images. Pixel values to be coded in the predictive coding schemes are predicted from already-coded adjacent pixels, and then, an entropy coder encodes the resulting prediction residuals. This paper describes an estimate and a selection method of the simple predictors in a theoretical viewpoint, which constrained as follows: (1) The predicted values are computed by weighted summation of the neighboring three pixels identical to JPEG and JPEG-LS; (2) The predicted values are obtained with a small number of shift operations, additions, and/or subtractions; (3) The predictor coefficients sum up to 1. The comparison of the total squared error between the simple predictors and the optimum predictors obtained by solving the normal equations is discussed. With the results of the theoretical analysis, we propose a six-predictor and an eleven-predictor set predictors. The performance of the proposed predictors on actual images is also presented.

This paper presents a block based lossless compression approach for gray scale images. Multiple scanning methods are applied to each block, and a newly proposed adaptive linear prediction is performed There are different prediction residuals obtained corresponding to different context based on multiple scanning. We choose the best residual for coding. That is, rather than relying any single scanning, our approach is to select a scanning produces the best result on each black. The prediction coefficients are updated during the scanning to optimize the coding accuracy. Experiment results show that our method out-performed JPEG-LS 4similar to5% in compression efficiency.

This paper presents an efficient lossless compression approach for gray scale images. The main contributions are 1) we divide the input image to blocks of pixels and use different scanning methods, we select the one that minimizes the prediction error for coding; 2) a new adaptive linear prediction is applied. Experiment results show that our method outperformed JPEG-LS 2 similar to 10% in compression efficiency.

In case that we separate RGB color still images into R-, G- and B- color planes, it is known that the color plane images have highly correlation each other. It is also known that the images have different properties in a locally called context. In this paper, we propose lossless compression method for RGB color still images which is realized by removing these redundancies. In the proposed method, we generate color difference component of the prediction errors obtained from predictive transformation on each plane. Then we separate the prediction errors and the color differences of the prediction errors according to their context, and apply optimal entropy coding for each context. From the simulation results, we confirmed that the compression of the proposed method improved by 14% and by 13% in comparison with that of the LOCO-I and that of the CALIC, respectively, and was also equivalent to that of the CREW. And we also confirmed that the processing time of the proposed method was faster than that of these conventional methods. © 2002, The Institute of Image Electronics Engineers of Japan. All rights reserved.

In video processing, compression techniques are required to improve the efficiency of transmission and storage of video data. Data compression methods can be classified into two basically different categories : lossy compression methods and lossless compression methods. Lossless compression methods are required for medical image processing and for compressions broadcast materials. We developed a novel video lossless compression method using the Hilbert curve, by mapping the three-dimensional data to one-dimensional data along the curve and then applying adaptive linear prediction coding to the one-dimensional data. Experimental results show that better compression ratio is obtained with our method than with lossless JPEG, although the processing time for both methods is almost equal.

The wide usage of small satellite imagery, especially its commercialization, makes data-based onboard compression not only meaningful but also necessary in order to solve the bottleneck between the huge volume of data generated onboard and the very limited downlink bandwidth. In this paper, we propose a method that encodes different regions with different algorithms. We use three shape-adaptive image compression algorithms as the candidates. The first one is a JPEG-based algorithm, the second one is based on the object-based wavelet transform (OWT) method proposed by [1], and the third adopts Hilbert scanning of the regions of interest followed by one-dimensional (1-D) wavelet transform. The three algorithms are also applied to the full image so that we can compare their performance on a whole rectangular image. We use eight Landsat Th I multispectral images and another 12 small satellite single-band images as our data set. The results show that these compression algorithms have significantly different performance for different regions. For relatively smooth regions, e.g., regions that consist of a single type of vegetation or water areas etc, the 1-D wavelet method is the best. For highly textured regions, e.g., urban areas, mountain areas, and so on, the modified OWT method wins over the others. For the whole image, OWT working at whole image mode, which is just an ordinary 2-D wavelet compression, is the heat. Based on this, we propose a new data-based compression architecture that extracts particular regions according to the application of interest and then involves different algorithms to encode different regions in order to achieve better performance than traditional onboard compression schemes in which a fixed compression method is applied to the whole image no matter what the application is. This approach is most appropriate for use with images captured by microsatellites, which are commissioned for specific applications in which one knows a priori which class of region the user is interested in.

A novel algorithm for complexity reduction in binary image processing, namely for computation of correlation between image and object template is proposed. This algorithm is based on direct computation of vector-matrix multiplication with utilisation of binary matrix factorisation approach. Comparison with other algorithms is given and it is shown that our approach allows to reduce tithe and complexity of this task.

Hilbert curve is a one-to-one mapping between N- dimensional (N-D) space and 1-D space. The Hilbert curve has been applied to image processing as a scanning technique (Hilbert Scan). Recently applications to multi-dimensional image processing are also studied. In this application, we use N-D Hilbert scan which maps N-D data to 1-D data along N-D Hilbert curve. However, N-D Hilbert scan is the application limited to data. in a hyper-cube region. In this paper, we present a novel algorithm for generating N-D pseudo-Hilbert curves in a hyper-rectangular parallelepiped region. Our algorithm is suitable for real-time processing and easy to implement, in hardware, since it is a simple and non-recursive. computation using look-up tables.

G. Peano has published a paper of space-filling curve in 1890. There are several applications using this curve in the area of image processing, computer graphics, etc. This paper presents a region-based scanning technique for image compression which we call an adaptive space filling scan (ASFS). In order to generate the ASFS, we make use of a minimum spanning tree technique. From several experiments for image compression, it is confirmed that in comparison to JPEG, acceptable quality images can be obtained at the same bit-rates.

We investigate a novel video compression method using the Hilbert curve. This curve is applied to various problems such as image compression, the traveling salesman problem, and database management, because of its locality preservation property. We map the three-dimensional data to one-dimensional data along the Hilbert curve and then apply a lossy compression method using a linear approximation to the one-dimensional data. This method uses simple segmentation of the one-dimensional data and does not require complex computation such as DCT or motion estimation. Experimental results show that our method obtains acceptable quality of reconstructed images comparable to H.263 at low bit rates but is about ten times faster than H.263 (full search motion estimation). It is suitable for video telephones which require real-time encoding of video signals.