Image and Graphics

Isometric shapes share the same geometric structure, and all possible bendings of a given surface are considered to have the same isometric deformation. Therefore, we use the inner distance to describe the isometric geometric structure. The inner distance is defined as the length of the shortest path between landmark points with the bending stability. Stochastic neighbor embedding algorithm t-SNE is a manifold embedding algorithm to visualize high-dimensional data by giving each data point a location in a two or three-dimensional map. Then, t-SNE is applied to 3D shapes isometric deformation in which Euclidean distances in high-dimensional space are replaced by inner distances. We can use this isometric deformation to describe invariant Signatures of surfaces, so that the matching of nonrigid shapes is better.

The issue of visual discomfort has long been restricting the development of advanced stereoscopic 3D video technology. Bolstered by the requirement of highly comfortable three-dimensional (3D) content service, predicting the degree of visual comfort automatically with high accuracy has become a topic of intense study. This paper presents a novel visual comfort assessment (VCA) metric based on sparse coding strategy. The proposed VCA metric comprises three stages: feature representation, dictionary construction, sparse coding, and pooling strategy, respectively. In the feature representation stage, visual saliency labeled disparity statistics and neural activities are computed to capture the overall degree of visual comfort for a certain stereoscopic image. A set of stereoscopic images with a wide range degree of visual comfort are selected to construct dictionary for sparse coding. Given an input stereoscopic image, by representing features in the constructed dictionary via sparse coding algorithm, the corresponding visual comfort score can be estimated by weighting mean opinion scores (MOSs) using the sparse coding coefficients. In addition, we conduct a new 3D image benchmark database for performance validation. Experimental results on this database demonstrate that the proposed metric outperforms some representative VCA metrics in the regard of consisting with human subjective judgment.

License plate extraction is an important step of License Plate Recognition (LPR) in Intelligent Transportation System. This paper presents a hybrid license plate extraction method which combines edge detection and support vector machine (SVM) filtering. Observing that there are many vertical edges in the license plate region, we firstly extract several candidate license plates according to vertical edge density, after that these candidate license plates are verified by a SVM classifier based on Histograms of Oriented Gradient texture descriptor (T-HOG). Promising results are achieved in the experiments.

A variety of saliency detection methods have been proposed in recently, which often complement each other. In this study, we try to improve their performances by aggregating these individual ones. First, we propose an improved Bayes aggregation method with double thresholds. Then, we compare it with five other aggregation approaches on four benchmark datasets. Experiments show that all the aggregation methods significantly outperform each individual one. Among these aggregation methods, average and Non-negative Matrix Factorization (NMF) weights perform best in terms of precision-recall curve, our Bayes is very close to them. While for mean absolute error score, NMF and our Bayes perform best. We also find that it is possible to further improve their performance by using more accurate reference map. The ideal is ground truth, of course. Our results could have an important impact for applications required robust and uniform saliency maps.

We propose a new digital watermarking method for the anti-counterfeit of trademarks based on complete orthogonal U system. By designing an algorithm and analyzing experimental results, we find that the proposed method is easily realizable and fights attacks like cutting, daubing, compression of JPEG and noise–adding. It is suitable for anti-counterfeit and authentication of trademark.

Iris segmentation is a vital forepart module in iris recognition because it isolates the valid image region used for subsequent processing such as feature extraction. Traditional iris segmentation methods often involve an exhaustive search in a certain large parameter space, which is time consuming and sensitive to noise. Compared to traditional methods, this paper presents a novel algorithm for accurate and fast iris segmentation. A gray histogram-based adaptive threshold is used to generate a binary image, followed by connected component analysis, and rough pupil is separated. Then a strategy of RANSAC (Random sample consensus) is adopted to refine the pupil boundary. We present Valley Location of Radius-Gray Distribution (VLRGD) to detect the weak iris outer boundary and fit the edge. Experimental results on the popular iris database CASIA-Iris V4-Lamp demonstrate that the proposed approach is accurate and efficient.

Edge detection in image processing is a difficult but meaningful problem. For noisy images, developing fast algorithms with good accuracy and stability is in need. In this work, we propose a variational model which links to the well-known Mumford-Shah functional and design a fast proximity algorithm to solve this new model through a binary labeling processing. Comparing with the famous Ambrosio-Tortorelli model, the efficiency and accuracy of the new model and the proposed minimization algorithm are demonstrated.

Foreground extraction is a fundamental step of video analysis. The common solution for foreground extraction is background subtraction which is based on color information. However, color is sensitive to intensity changes and may lose efficacy in complex scenes such as scene with low contrast or strong illumination. To overcome the disadvantages of color-based methods, we propose a new approach based on edge information. We get the edge of foreground from color-based background instead of edge-based background to reduce calculation amount. And a novel multi-resolution edge aggregation algorithm is used to solve the edge-filling problem, especially in the case when edge is not continuous. This algorithm obtains foreground region through expands the influence of the edges and reduces the gaps between the edges. Both visual and quantitative comparisons on various image sequences validate the efficacy of our method.

Image blur caused by camera shake is often spatially variant, which makes it more challenging to recover the latent sharp image. Geometrical camera shake model based non-uniform deblurring methods, modeling the blurry image as a weighted summation of the homographically transformed images of the latent sharp image, although can achieve satisfactory deblurring results, still suffer from the problems of heavy computation or extensive memory cost. In this paper, we propose a generalized additive convolution (GAC) model for efficient non-uniform deblurring. A camera motion trajectory can be decomposed into a compact set of in-plane translations (slice) and roll rotations (fiber), which with an insightful analysis can both be formulated as convolution. The GAC model provides a promising way to overcome the difficulties of both computational load and memory burden. The experimental results show that GAC can obtain satisfactory deblurring results, and is much more efficient than state-of-the-arts.

High dynamic range (HDR) imaging technology has been widely implemented in digital microscopes for taking still images of high contrast specimens. However, capturing HDR microscopic video is much more challenging. In this paper, we present a HDR microscopic video system based on GPU accelerated computing. We show that by combining CPU and GPU computing, it is possible to build a stable HDR video system using a single off the shelf camera. We show that capturing multiple frames of different exposure intervals, aligning consecutive neighboring frames, constructing HDR radiance map and tone mapping the radiance map for display, can be realized by using GPU computing to accelerate the processing speed. We present experimental results to show the effectiveness of our system and how HDR video can reveal much more details than conventional videos.

In this paper, we propose a new method for learning a No-Reference Image Quality Assessment that relies on the preparation of examples of different classes of distorted images. Our learning model learns classes of quality for each distortion type. This is achieved by preparing classes of training examples in a way that for each type of distortion, while we introduce a level of distortion to go from “high quality” to “low quality” care must be taken in that a lower rate (integer) is given to the new class for which the actual degree of the distortion have degraded the perceptual quality. A radial basis function (RBF) network is trained using some extracted perceptual features to separate the images into ten categories of quality. The same RBF reuses the same features to quantify a distorted test image by predicting its category of quality. Experimental results on the Distorted Face Database show that the method is effective.

Dimensionality reduction is widely used in image understanding and machine learning tasks. Among these dimensionality reduction methods such as LLE, Isomap, etc., PCA is a powerful and efficient approach to obtain the linear low dimensional space embedded in the original high dimensional space. Furthermore, Kernel PCA (KPCA) is proposed to capture the nonlinear structure of the data in the projected space using “

Kernel Trick

”. However, KPCA fails to consider the locality preserving constraint which requires the neighboring points nearer in the reduced space. The locality constraint is natural and reasonable and thus can be incorporated into KPCA to improve the performance. In this paper, a novel method, which is called Locality Preserving Kernel PCA (LPKPCA) is proposed to reduce the reconstruction error and preserve the neighborhood relationship simultaneously. We formulate the objective function and solve it mathematically to derive the analytical solution. Several datasets have been used to compare the performance of KPCA and our novel LPKPCA including ORL face dataset, Yale Face Dataset B and Scene 15 Dataset. All the experimental results show that our method can achieve better performance on these datasets.

A method of overcoming occlusion in vehicle tracking system is presented here. Firstly, the features of moving vehicles are extracted by the vehicle detection method which combines background subtraction and bidirectional difference multiplication algorithm. Then, the affection of vehicle cast shadow is reduced by the tail lights detection. Finally, a two-level framework is proposed to handle the vehicle occlusion which are NP level (No or Partial level) and SF level (Serious or Full level). On the NP level, the vehicles are tracked by mean shift algorithm. On the SF level, occlusion masks are adaptively created, and the occluded vehicles are tracked in both the original images and the occlusion masks by utilizing the occlusion reasoning model. The proposed NP level and SF level are sequentially implemented in this system. The experimental results show that this method can effectively deal with tracking vehicles under ambient occlusion.

By using facial motion data from passive optical motion capture system, this paper proposed an animation reconstruction approach, which is aimed to solve the common problem: model discontinuity and reconstructing a real-like facial animation. For this problem, a facial animation reconstruction approach with model divisions is proposed. We first divide facial model into different regions and request the deformed motion data by deploying RBF mapping. Then we adjust the deformed motion data to ensure their exactness. Besides, considering the nonlinear relationship of neighbors with the current point, we also proposed a facial animation reconstruction approach without model divisions, where we apply the motion of one-ring neighbors to adjust the deformed motion data. The experiments show that the proposed methods have an acceptable precision of reconstruction.

Shape-based image recognition is a key technology in computer vision, and Fourier descriptor (FD) is one important way to describe such images. FD uses the Fourier Transform of the contour coordinate as eigenvector to describe the image contour property. However, it can only show the contour property, but may fail to distinguish images with the same contour but different content. And the length of FD varies with the size of the image. In this paper, a modified Fourier descriptor (

MFD)

is proposed, which is invariant with translation, rotation, and scaling of the image. It takes both the contour property and the content into consideration. A 2D shape-based image can be represented by a one-dimensional discrete array with constant length, which makes it convenient to recognize different images. To prove the efficiency of the proposed algorithm, we have applied it to shape recognition experiments and got reasonable results.

Event recognition in the consumer videos is a challenging task since it is difficult to collect a large number of labeled training videos. In this paper, we propose a novel Multiple kernel Image Group Adaptation approach to divide the training labeled Web images into several semantic groups and optimize the combinations of each based kernel. Our method simultaneously learns a kernel function and a robust Support Vector Regression (SVR) classifier by minimizing both the structure risk of SVR with the smooth assumption and the distribution difference of weighted image groups and the consumer videos. Comprehensive experiments on the datasets CCV and TREATED 2014 demonstrate the effectiveness of our method for event recognition.

High accuracy and real-time implementation are important requirements for scene matching in visual navigation applications. A multi-scale coarse-to-fine SAR scene matching algorithm based on compressed contour feature is proposed in this paper. Firstly, the compressed contour features of the real-time image and the reference image are extracted through adjacent sub-region differencing, and multi-scale feature images are constructed by changing sub-region size. Then, coarse matching is carried out on the feature images, and scale factor and coarse matching position are obtained through cluster analysis of several matching results. Finally, some sub-regions with rich edge information are chosen from the original real-time image, and used to carry out fine matching around the coarse matching region in the original reference image, in this way the final accurate matching position is obtained. Experimental results demonstrate the excellent matching performance of the proposed algorithm.

Impact craters are typical lunar areas which can reflect the characteristics of lunar surface, so the studies of them are one of the key tasks of lunar exploration. A class of complete orthogonal piecewise polynomials in L

2

[0,1] called V-system is introduced in this paper and this new approach can be used to represent the contour lines of lunar DEM data. It is not only introduced for accurately representing the contour lines but also eliminating effectively the Gibbs phenomenon. Based on V-system, there is an algorithm for transferring a given contour to V-spectrum. The proposed algorithm is intuitive, easy and fast. Some examples of lunar contour maps’ representation in V-system are given.

Aiming to saliency detection problem of degraded foggy images, a new method of object saliency detection method in foggy images based on region covariance matrix is presented. In the method, color, direction and space information are extracted to form covariance feature description matrix according to characteristics of foggy images. Then local saliency sub-map is acquired by local contrast. In the same time, Wiener filter and Sobel edge detection are used to obtain global saliency sub-map. Finally, local saliency map of color domain is optimized by edge global saliency map, and saliency map is completed. Experiments show that compared with state-of-art methods, the proposed method has better adaptability and accuracy to object saliency detection in foggy images.

In this paper, we propose a new optimization algorithm utilizing the texture feature and correlation of adjacent frames for newly proposed video standard High Efficiency Video Coding (HEVC). For intra prediction, the complexity of pictures’ texture to perform different levels of simplification on Most Probable Mode (MPM) selection is scaled. For inter prediction, current Coding Unit (CU) depth information with that information of temporally adjacent frame’s co-located CU is initialized. Experimental results show that the proposed algorithm improves the efficiency of encoder with more than 30 percent of encoding time decreaseand nearly negligible increment in bit-rate.

As an important remote sensing image pre-processing method, radiation correction is essential to reduce deviation introduced by environment factors, especially for tasks such as image compression, image fusion, and object recognition. In this paper, we propose a new radiation correction method for remote sensing images based on change detection. Due to the fact that areas of with significant changes deteriorate performance of radiation correction, these sections should be detected and discarded in the image firstly. Then general radiation correction technology is considered to have better performance on the rest sections. The core idea of the proposed method exists in the combination of radiation correction and change detection. Experimental results prove validness of the proposed method. As an application example, this method used in image compression shows better performance than other compression technologies.

In order to overcome the shortcomings of existing maniflod learning algorithm, a new supervised manifold algorithm, which improves the original algorithm and makes it more reasonably, has been proposed. Firstly, a more accurate within-class scatter matrix only with the samples belong to the same class is established to characterize the local structure of each manifold. Secondly, nearby maniflods, which can reflect the relationships of different maniflods more accurately, are selected to establish the between-class scatter matrix to characterize the discreteness of different maniflods. Finally, the Fisher criterion is used to solve the objective function and get the optimal projection direction, which can maximize the ratio of the trace of the between-class scatter matrix to the trace of the within-class scatter matrix. Experimental results demonstrate that the proposed algorithm is effective in feature extraction, leading to promising recognition performance in face recognition.

Accurate brain image segmentation is a challenge and meaningful task that assists physicians in the disease diagnosis. In this paper, we present a nonlocal based two-step method for image segmentation. First step is to denoise the

MRI

brain image with adaptive nonlocal regularization. The second step is our new nonlocal based regularized segmentation. We force the result segmentation of grey matter(

GM

), white matter(

WM

) and cerebrospinal fluid(

CSF

) keeping as much structure as possible by using nonlocal regularization, which has significant meaning in diagnosis. With experiments on synthetic images from

BrainWeb

and real

MRI

images from Zhejiang Cancer Hospital, we find that our method performances very well on both databases.

One of the indispensable prerequisites for high resolution remote sensing image interpretation and processing is successful image segmentation. The algorithm presented in this paper aims for a high efficient image segmentation applicable and adaptable to high resolution remote sensing images. This is achieved by a non-seed-based region growing, which constructs neighbor pairwise pixel stack instead of depending on any seed points. The stack is constructed in increasing order of neighbor pairwise pixel spectral difference which is computed based on 4-connexity. The proposed algorithm carries out region growing according to the merging criterion (i.e. grow formula) and traversal of the stack. We apply the proposed and conventional region growing algorithms to two data sets of ZiYuan-3 (ZY-3) high resolution remote sensing images and analyze the segmentation results based on Carleer evaluation method that manifests high efficient segmentation of the proposed algorithm.

Extracting well-distributed and precisely aligned control points (CPs) is extremely important for remote sensing image registration, particularly for high resolution images with large local distortion. Based on a theoretical analysis of estimation perturbation in transformation parameters, a novel CP dispersion approach is proposed to select high quality and uniformly distributed CPs. This approach retains a minimum spanning tree (MST) of the selected CPs during the algorithm and adds a new CP to the tree in each iteration until satisfying the convergence condition. Moreover, to acquire adequate number of CPs for the dispersion process, a coarse-to-fine matching approach is proposed. Experiment results indicate that the proposed method improves the match performance compared to other CP dispersion methods in terms of aligning accuracy.

In this paper, we propose a novel framework of Foreign Object Debris (FOD) classification system. The system contains a FOD detection subsystem, electro-optical subsystem and the control center. The system not only provides continuous surveillance of scanned surfaces and achieves the goal of FOD detection, but also performs FOD classification. Both low level features and subspace features are compared to extract the FOD. Multiclass classifiers are trained in all the candidate feature spaces with the Support Vector Machine (SVM) to classify FOD. Experimental results show that it is promising to classify FOD with low-level features.

This paper presents a novel infrared (IR) and visible images fusion methodology based on non-subsampled contourlet transform (NSCT). NSCT shows better performance compared with usual multi-scale decomposition for its multi-scale, shift invariance, multi-direction and efficient capture of geometric structures. The proposed fusion method uses NSCT for multiresolution decomposition of the source images. The low-pass NSCT adaptive fusion weights calculated from the IR source image’s pixel statistical characteristics. The high frequency directional coefficients with max absolute value are the coefficients of the fusion NSCT high frequency. Experimental results conforms that the proposed method have better performance compared with DWT, compressed sensing based on DWT (CS-DWT), NSCT, NSCT based on spatial frequency motivated pulse coupled neural networks (SF-PCNN-NSCT) from visual effects and a list of fusion quality evaluation metrics.

In this paper, we propose a novel improved binarized normed gradients (BING) objectness method based on the multi-feature boosting learning. A series of difference of gaussians (DoG) of the images with given parameters are used for the feature extraction stage, since the image DoG filter can better describe objects border. In addition, in training phase, the classifier can adaptively combine the features from different scales and different frequency components. Moreover, since the norm of the feature gradients is a simple 64D feature in the proposed framework, the computational complexity of the algorithm is in the same level compared with the BING measure. Experiments on the challenging PASCAL VOC 07 dataset show that the proposed method can not only achieve higher detection rate and average score than some current related objectness measures, but also lead to a very competitive accuracy of locating objects, even in some difficult cases.

Human identification has been a prominent area in the field of computer vision and artificial intelligence. In this paper, a novel human identification method is proposed which is based on a Cubic Bezier Curve (CBC) and statistical techniques through 3D joint movement data. Data acquisition from motion capture system that provides accurate motion information of body joints in 3D environment. Such type of data has sole properties which distinguishes between images and videos. The simple kalman filter that can be used for removing noise in data, is guided by smooth and compactness manners. The features of the human body joints one upper joint (shoulder) and three lower joints (hip, knee and ankle) are computed by using the statistical moments. These features are used as the control points of the curve. The curve passes through the control points, which describes the relationship among joints muscles in human walking. Statistical techniques are applied to CBC coordinates for human recognition. Here, the rotation angle data of the joints is extracted from Biovison Hierarchical data, because these four joints provide the discriminating confusion of deduced information of human joints for identification through gait curve. The performance of our method is evaluated on CMU database. It achieves 100 % accuracy rate of identification by using the proposed database.

This paper presents a novel method for barcode detection and recognition in the scenes consisting of multiple barcodes. In order to effectively detect the locations of the barcodes, we construct a new feature parameter, the ratio of horizontal gradient to vertical gradient of each sub-block of the image, which can reflect the distinctive feature of barcode area from other parts of the image. We then use this feature to detect the position of each barcode with an adaptive threshold. Furthermore, in considering that the information of a barcode is encoded in the widths of bars, we develop a method for extracting the width information by reconstructing a high-resolution binary barcode image. The reconstruction is realized through projecting barcode image along optimal directions and interpolating the projection curve by cubic spline interpolation method. Experiment results show that the proposed approach is effective in detecting and recognizing multiple barcodes in complex scenes.

Facial beauty prediction is a challenging task in pattern recognition and biometric recognition as its indefinite evaluation criterion, compared with the other facial analysis task such as emotion recognition and gender classification. There are many methods designed for facial beauty prediction, whereas they have some limitations. Firstly, the results are almost achieved on a relative small-scale database, thus it is difficult to model the structure information for facial beauty. Secondly, most facial beauty prediction algorithm presented previously needs burdensome landmark or expensive optimization procedure. To this end, we establish a larger database and present a novel method to predict facial beauty. The works in this paper are notably superior to previous works in the following aspects: (1) A large database is established whose distribution is more reasonable and utilized in our experiments; (2) Both female and male facial beauty are analyzed under unconstrained conditions without landmark; (3) Multi-scale apparent features are learned by our method to represent facial beauty which is more expressive and requires less computation expenditure. Experimental results demonstrate the efficacy of the presented method from the aspect of accuracy and efficiency.

For the calculation complexity problem of image fusion based on traditional non-subsampled contourlet transform (NSCT), a novel NSCT fusion algorithm based on compressive sensing (CS) and dynamic weighted non-negative matrix factorization (DWNMF) is proposed. Firstly, NSCT is employed to decompose source images, for band-pass directional sub-band coefficients which are featured with high calculation complexity, CS theory is applied to fuse them, meanwhile, the fusion approach based on DWNMF is employed to fuse low-pass sub-band coefficients. Secondly, the minimum total variation (min-TV) algorithm is adopted to reconstruct fused band-pass directional sub-band coefficients in CS domain, to get fused band-pass sub-band coefficients in NSCT domain. Finally, using the inverse NSCT transform to reconstruct fused low-pass and band-pass sub-images, to gain the final fused image. Simulation results show that, the proposed approach can not only improve the fusion effect, but also reduce the calculation complexity.

This paper proposes a novel prediction algorithm in Gaussian mixture probability hypothesis density filter for target tracking in linear dynamical model. In tracking algorithms, with a possibility of multiple measurements per target, a model for the number of measurements per target is needed. Lately, different implementations have been proposed for such targets. To do a better estimation of performance, this work relaxes the Poisson assumptions of target tracking probability hypothesis density filter in targets and measurement numbers. We offered a gamma Gaussian mixture implementation capable of estimating the measurement rates and the kinematic state of the target. The Variational Bayesian approximation converts the Gamma-Gaussian mixture into the improved Gaussian mixture with its news mean and covariance components. It is compared to its GM-PHD filter counterpart in the simulation study and the results clearly show the best performance of the improved Gaussian Mixture algorithm.

The point inside/outside a polygon test is used by many applications in computer graphics, computer games and geographical information systems. When this test is repeated several times with the same polygon a data structure is necessary in order to reduce the linear time needed to obtain an inclusion result. In the literature, different approaches, like grids or quadtrees, have been proposed for reducing the complexity of these algorithms. We propose a new method using a polar space subdivision to reduce the time necessary for this inclusion test. The proposed algorithm is robust and has a performance of

$$ O(k) $$

, where

$$ k \ll N $$

,

$$ k $$

is the number of tested intersections with polygon edges, and the time complexity of the preprocessing is

$$ O(N) $$

, where

$$ N $$

is the number of polygon edges.

The speed of Forest fire detection is important for fire safety. This paper presents a forest fire detection algorithm. At first the images were compressed through combining DCT code and RLE code, then the suspicious images were priority identified at the decoder, at last color identification was detected using HSI color space model flame. Experimental result shows that this algorithm can quickly achieve rapid identification with high accuracy. It has a strong anti-jamming capability and application prospects.

In the digital multimedia era, data hiding in the compression domain becomes increasingly popular in the need for speeding up the transmission process and reducing bandwidth. Recently, VQ-based watermarking techniques have attracted more attentions, e.g., Tu and Wang proposed a VQ-based lossless data hiding scheme with high payload most recently. However, their scheme produces some more overhead information. In this study, we develop a novel reversible hiding scheme which may reduce the use of overhead bits, which is especially effective for images of complex texture. Specifically, a codebook is partitioned into six clusters which are organized based on the usage frequency of codewords. We then develop a new mapping relationship among the six clusters to hide secret data more cost-effectively. The experimental results demonstrate that our proposed scheme outperforms Tu’s scheme for complex texture images.

In traffic surveillance system, it is still a challenging issue to track an occluded vehicle continuously and accurately, especially under total occlusion situations. Occlusion judgment is critical in occluded target tracking. An occlusion judgment scheme with joint parameters is proposed for target tracking method based on particle filter. A corner matching method is utilized to improve the accuracy of target position and velocity estimation due to structure information, thus obtain a more accurate weight value which can reflect the real target’s status. By analyzing the internal relation between the weight value and the particles distribution region based on resample function of particle filter, a new parameter with good performance is proposed to improve the occlusion detection efficiency.

The change of object’s scale is an important reason leading to tracking failure in visual tracking. A scale adaptive tracking algorithm based on kernel ridge regression and Fast Fourier Transform is proposed in this paper. Firstly, the algorithm build regression model using appearance information of object, and then get the position of object in the search region using the regression model. Finally, it estimates the best scale by considering the weight image of all pixels in the candidate region. The experimental results show that the proposed algorithm not only can track the object real time, but also adapt to the changing of object’s scale and the interference of background. Compared with the traditional ones, it owns good robustness and efficiency.

Automatic reconstruction of cross-cut shredded text documents (RCCSTD) is important in some areas and it is still a highly challenging problem so far. In this work, we propose a novel semi-automatic reconstruction solution archive for RCCSTD. This solution archive consists of five components, namely preprocessing, row clustering, error evaluation function (EEF), optimal reconstructing route searching and human mediation (HM). Specifically, a row clustering algorithm based on signal correlation coefficient and cross-correlation sequence, and an improved EEF based on gradient vector is separately evaluated by combining with HM and without HM. Experimental results show that row clustering is effective for identifying and grouping shreds belonging to a same row of text documents. The EEF proposed in this work improves the precision and produces high performance in RCCSTD regardless of using HM or not. Overall, extra HM boosts both of the performance of row clustering and shred reconstructing.

There are growing concerns on the potential side effect of radiation, which could be decreased by lowering the tube current. However, this manner will lead to a degraded image since X-ray imaging is a quantum accumulation process. Great efforts have been devoted to alleviate this problem. In this paper, a simple Wiener filter was employed to denoise projection data in detector domain. And then an enhancement filter was exploited to strengthen small structures. As a consequence, this combination leaded to a reconstruction with good trade-off between noise and resolution. For the purpose of comparison, block-matching and 3D/4D denoising algorithm (BM3D/BM4D) were also adopted to denoise the projections. Experimental results demonstrated that the proposed algorithm could deliver a reconstruction with comparable quality as BM4D algorithm while better than that of BM3D denoising algorithm. Note that the propose algorithm has a much higher computation efficiency than BM3D/BM4D, and hence provides an insight into the clinical utility where real time is of high importance.

In this paper, we propose a novel smooth approximation algorithm of rank-regularized optimization problem. Rank has been a popular candidate of regularization for image processing problems,especially for images with periodical textures. But the low-rank optimization is difficult to solve, because the rank is nonconvex and can’t be formulated in closed form. The most popular methods is to adopt the nuclear norm as the approximation of rank, but the optimization of nuclear norm is also hard, it’s time expensive as it needs computing the singular value decomposition at each iteration. In this paper, we propose a novel direct regularization method to solve the low-rank optimization. Contrast to the nuclear-norm approximation, a continuous approximation for rank regularization is proposed. The new method proposed in this paper is a ‘direct’ solver to the rank regularization, and it just need computing the singular value decomposition one time, so it’s more efficient. We analyze the choosing criteria of parameters, and propose an adaptive algorithm based on Morozov discrepancy principle. Finally, numerical experiments have been done to show the efficiency of our algorithm and the performance on applications of image denoising.

This paper presents a parameter-free document binarization method based on text characteristics. For a given stroke width, the text and background regions in binarized object regions are estimated with morphological operators. Then according to the numbers of the text and background pixels an optimal threshold is determined. To make the proposed method parameter-free, an automatic estimation of stroke width is also proposed based on the ratio of thick stroke pixels to binarized object pixels. Document images with various degenerations, such as stain and bleed-through, were used to test the proposed method. Comparison results demonstrate that our method can achieve a better performance than Otsu’s, Kwon’s, Chen’s, Transition, and Moghaddam’s binarization methods.

In recent years, deformation transformation techniques have been applied to the recognition, object matching, and image correction of Chinese characters. In this study, we propose a novel deformation transformation approach that can change the style of printed Chinese characters (PCCs). We believe that each category of PCC has the same topological structure and can be considered a type of deformable object. Therefore, we can change the style of each PCC through some suitable deformations. To this end, we propose a 1-D deformable transformation method that can deform PCCs into different styles. Such a style transformation method will not only enrich Chinese character fonts, but also enlarge a given database so as to represent more variations than the original.

Generally, for a global or local color transfer, the traditional approaches will rearrange color distribution in source image according to reference image. However, the destruction of scene detail and illumination environment might produce a low-fidelity color transfer result. In this paper, we propose a unified fidelity optimization model for color transfer to yield a high-fidelity transfer result in terms of color, detail and illumination. Corresponding to the three characteristics, our approach is described as an optimization problem with three energy terms: color mapping, detail preserving and illumination awareness. Color mapping can employ histogram matching to impose the color style of reference image on source image; Detail preserving can apply gradient guidance to maintain scene detail in source image; Illumination awareness can construct illumination affinity to harmonize illumination environment. Moreover, following the definition of fidelity with three characteristics, we also propose an objective evaluation metric to analyze the performance of our approach in different coefficients. The comparison of experiment results demonstrates the effectiveness of our optimization model.

In this paper, we develop a new vision-based parking space surveillance system for parking lot management. The system consists of three parts. Initially, a feature-based background model using edge and color characteristics is proposed, and foreground feature is extracted to determine whether the parking space is vacant or not. Secondary, to capture the pictures when a car has been completely into the parking space, we employ adjacent frame difference image to find the static state of the parking space. Finally, for the final decision, an adaptive thresholds updating method is proposed. After experiments on different parking lots, the proposed system has been proved to be effective and accurate.

Since video logos are embedded into complex background and normally hollow or translucent, traditional video logo recognition techniques are facing issues such as difficulties in extracting image features, high complexity of matching algorithm and feature dimension disaster. A well-behaved TV logo recognition method using heuristic optimality checked basis pursuit denoising is firstly proposed in the paper. Original pixels are directly used for recognition in this method, avoiding the difficulties in extracting image features as well as being able to recognize hollow and translucent logos with relatively better robustness. Experiments have shown that by applying this method, average logo recognition time is within 30 ms for a single-frame video with recognition accuracy more than 98 %.

We propose an automatically and accurately facial landmark localization algorithm based on Active Shape Model (ASM) and Gabor Wavelets Transformation (GWT), which can be applied to both 2D and 3D facial data. First, ASM is implemented to acquire landmarks’ coarse areas. Then similarity maps are obtained by calculating the similarity between sets of Gabor jets at initial coarse positions and sets of Gabor bunches modeled by its corresponding manually marked landmarks in the training set. The point with the maximum value in each similarity map is extracted as its final facial landmark location. It is showed in our laboratory databases and FRGC v2.0 that the algorithm could achieve accurate localization of facial landmarks with

state

-

of

-

the

-

art

accuracy and robustness.

Fiber tracking is a basic task in analyzing data obtained by diffusion tensor magnetic resonance imaging (DT-MRI). In order to get a better tracking result for crossing fibers with noise, an improved fiber tracking method is proposed in this paper. The method is based on the framework of Bayesian fiber tracking, but improves its ability to deal with crossing fibers, by introducing the high order tensor (HOT) model as well as a new fiber direction selection strategy. In this method, orientation distribution function is first obtained from HOT model, and then used as the likelihood probability to control fiber tracing. On this basis, the direction in candidates that has the smallest change relative to current two previous directions is selected as the next tracing direction. By this means, our method achieves better performance in processing crossing fibers.

Image registration is an essential technology for image guided diagnosis. However, in gastroscopic environment, it is still a challenging work due to the ambiguous and noisy endoscopic images. In this paper, we propose an iterative image registration method, homographic triangle and epipolar constraint registration, based on the homographic hypothesis. The method starts with establishing initial matching point pairs between gastroscopic image sequences and clustering them by Delaunay triangulation; normalized cross correlation is then introduced to validate the homographic assumptions of the matching triangles; after that, the inscribed circle’s center point of an unmatched triangle is registered by the epipolar constraint; finally, each unmatched triangles is divided into three sub-triangles according to its vertexes and inscribed circle center point for next HTECR iteration. Clinical data experimental results show a promising performance with this method.

This paper presents an improved algorithm to generating support structure for 3D printing model. Firstly, the candidate regions are obtained from the shadow area in each layer. Secondly, the maximal random points sampling in anchor map is implemented by the Poisson disk sampling. Thirdly, the support structure is generated in each layer according to the sampling points for the 3D model. The experimental results compared with state-of-the-art algorithm show that the proposed algorithm can save about 5–20 % support sticks.

In this paper, we propose an outlier removal method which utilizes the information of hypotheses for model fitting. The proposed method statistically analyzes the properties of data points in two groups of hypotheses, i.e., “good hypotheses” and “bad hypotheses”. We show that the bad hypotheses, whose parameters are far from the parameters of model instances in data, also contain the correlation information between data points. The information can be used to effectively remove outliers from the data. Experimental results show the proposed method can effectively remove outliers on real datasets.

The novel approach presented in this paper aims for unsupervised change detection applicable and adaptable to remote sensing images. This is achieved based on a combination of principal component analysis (PCA) and genetic algorithm (GA). The PCA is firstly applied to difference image to enhance the change information, and the significance index

F

is computed for selecting the principal components which contain predominant change information based on Gaussian mixture model. Then the unsupervised change detection is implemented and the resultant optimal binary change detection mask is obtained by minimizing a mean square error (MSE) based fitness function using GA. We apply the proposed and the state-of-the-art change detection methods to ASTER and QuickBird data sets, meanwhile the extensive quantitative and qualitative analysis of change detection results manifests the capability of the proposed approach to consistently produce promising results on both data sets without any priori assumptions.

The artificial vision certainly refers to image processing, these images are only the raw material of a much broader science, the same as strives to emulate human perceptual abilities, Sudoku is one of the most popular puzzle games of all time, for this reason was interesting to apply the programming knowledge to solve a common daily challenge, the goal of Sudoku is to fill a 9 × 9 grid with numbers so that each row, column and 3 × 3 section contain all of the digits between 1 and 9. The aim of the work is show that with a knowledge of programming with a webcam in a pc is possible to apply techniques of image processing to detect the Sudoku area and solve it. All the work was developed in C ++ using QTcreator Ide, OpenCV, Tesseract Libraries and the code to solve the Sudoku is open source.

Spectral-Domain Optical Coherence Tomography (SD-OCT) is a non-invasive imaging modality, which provides retinal structures with unprecedented detail in 3D. In this paper, we propose an automated segmentation method to detect intra-retinal layers in OCT images acquired from a high resolution SD-OCT Spectralis HRA+OCT (Heidelberg Engineering, Germany). The algorithm starts by removing all the OCT imaging artifects includes the speckle noise and enhancing the contrast between layers using both 3D nonlinear anisotropic and ellipsoid averaging filers. Eight boundaries of the retinal are detected by using a hybrid method which combines hysteresis thresholding method, level set method, multi-region continuous max-flow approaches. The segmentation results show that our method can effectively locate 8 surfaces for varying quality 3D macular images.

Segmentation of the liver from abdominal CT images is a prerequisite for computer aided diagnosis. However, it is still a challenging task due to the low contrast of adjacent organs and the varying shapes between subjects. In this paper, we present a liver segmentation framework based on prior model using level set. We first weight all of the atlases in the training volumes by calculating the similarities between the atlases and the test image to generate a subject-specific probabilistic atlas. Based on the generated atlas, the most likely liver region (MLLR) of the test image is determined. Then, a rough segmentation is performed by a maximum a posteriori classification of probability map. The final result is obtained by applying a shape-intensity prior level set inside the MLLR with narrowband. We use 15 CT images as training samples, and 15 exams for evaluation. Experimental results show that our method can be good enough to replace the time-consuming and tedious manual approach.

3D character with human motion offers a high end technology for creating compelling contents in graphics. In this paper, we present an automatic system to animate a 3D character with human motion streamed from a single video camera. The traditional mesh animation process is laborious and requires high skills from the users. To mitigate this limitation, a new way for bringing 3D objects to life is introduced that does not need explicit mesh positioning. In our framework, the animation is driven by the captured motion from an ordinary RGB camera. In order to reduce the ambiguity of the estimated 3D pose, a modified spatio-temporal constraint based algorithm is used for articulated gesture estimation across frames while maintaining temporal coherence. Our approach demonstrates promising performance on par with state-of-the-art techniques. We believe the presented animation system will allow a new audience of novice users to easily and efficiently create animation for arbitrary 3D characters.

Brain lesions, especially White Matter Lesions, are not only associated with cardiac and vascular disease, but also with normal aging. Quantitative analysis of WMLs in large clinical trials is becoming more and more important. Based on intensity features and tissues’ prior, we develop a computer-assisted WMLs segmentation method, with the expectation that our approach would segment WMLs in Magnetic Resonance Imaging (MRI) sequences without user intervention. We first train a SVM nonlinear classifier to classify the MRI data voxel-by-voxel. In detail, the attribute vector is constructed by the intensity features extracted from Multi-channel MRI sequences, i.e. fluid attenuation inversion recovery (FLAIR), T1-weighted, T2-weighted, proton density-weighted (PD), and the tissues’ prior provided by partial volume estimate (PVE) images in native space. Based on the prior that the lesions almost exist in white matter, we then present an algorithm to eliminate the false-positive labels. Subsequent further segmentation through Region-Scalable Fitting (RSF) evolution on FLAIR sequence is employed to effectively segment precise lesions boundary and detect missing lesions. Compared with the manual segmentation results from an experienced neuroradiologist, experimental results for real images show desirable performances and high accuracy of the proposed method.

This paper seeks to develop fast and accurate endoscopic stereo 3-D scene reconstruction for image-guided robotic surgery. Although stereo 3-D reconstruction techniques have been widely discussed over the last few decades, they still remain challenging for endoscopic stereo images with photometric variations, noise, and specularities. To address these limitations, we propose a robust stereo matching framework that constructs cost function on the basis of image gradient and three-channel color information for endoscopic stereo scene 3-D reconstruction. Color information is powerful for textureless stereo pairs and gradient is robust to texture structures under noise and illumination change. We evaluate our stereo matching framework on clinical patient stereoscopic endoscopic sequence data. Experimental results demonstrate that our approach significantly outperforms current available methods. In particular, our framework provided 99.5 % reconstructed density of stereo images compared to other available matching strategies which achieved at the most an 87.6 % reconstruction of the scene.