This PDF file contains the editorial “JEI 2014 List of Reviewers” for JEI Vol. 24 Issue 02

Image deblurring is one of the classical problems in image processing. The main purpose of image deblurring is to restore the images corrupted by blur and additive noise while preserving edges and details. We focus on the nonblind image deblurring (NBID) problem. In order to obtain good deblurred images, we propose an NBID method based on a compound regularization model associated with the shearlet-based sparsity and the weighted anisotropic total variation. Based on this model, we present an alternative iterative scheme which consists of the operator splitting and penalty techniques. The split Bregman-based multivariable minimization iteration is introduced to optimize the proposed NBID inverse problem. Compared with some previous methods, the experiments demonstrate the efficiency and viability of the proposed method for preserving sharp edges and structural details of the image while mitigating the artifacts.

In the quantization-based watermarking framework, the perceptual just noticeable distortion (JND) model has been widely used to determine the quantization step size, as it can provide a better tradeoff between fidelity and robustness. However, the calculated JND values can vary due to changes introduced by watermark embedding. As a result, the mismatch problem will lead to watermark extraction errors in the absence of attacks. We present an improved spread transform dither modulation (STDM) watermarking scheme. Performance improvement with respect to the existing algorithm is obtained by a discrete cosine transform (DCT)-based perceptual JND model that is highly compatible with the STDM watermarking algorithm. The proposed scheme not only incorporates various masking effects of human visual perception, but also avoids the mismatch problem by utilizing a new measurement of the pixel intensity and edge strength. In contrast to conventional JND models, the proposed model can be theoretically invariant to the changes in the watermark-embedding processing, therefore, more fit for quantization-based watermarking. Experimental results confirm the improved robustness performance of the JND model in the STDM watermarking framework. Simulation results show that the proposed scheme is more robust than the existing JND model-based watermarking algorithms with uniform fidelity. Furthermore, our proposed scheme has a superior performance compared with previously proposed perceptual STDM schemes.

We present a three-dimensional (3-D) reconstruction system designed to support various autonomous navigation applications. The system presented focuses on the 3-D reconstruction of a scene using only a single moving camera. Utilizing video frames captured at different points in time allows us to determine the depths of a scene. In this way, the system can be used to construct a point-cloud model of its unknown surroundings. We present the step-by-step methodology and analysis used in developing the 3-D reconstruction technique. We present a reconstruction framework that generates a primitive point cloud, which is computed based on feature matching and depth triangulation analysis. To populate the reconstruction, we utilized optical flow features to create an extremely dense representation model. With the third algorithmic modification, we introduce the addition of the preprocessing step of nonlinear single-image super resolution. With this addition, the depth accuracy of the point cloud, which relies on precise disparity measurement, has significantly increased. Our final contribution is an additional postprocessing step designed to filter noise points and mismatched features unveiling the complete dense point-cloud representation (DPR) technique. We measure the success of DPR by evaluating the visual appeal, density, accuracy, and computational expense and compare with two state-of-the-art techniques.

We propose an effective physical model based single-image dehazing algorithm under the regularization framework. Using an anisotropic Laplacian term as a constraint, the atmospheric propagation model that encodes the thickness of the haze is estimated by the regularization theory. Employing the physical model, haze-free images are reconstructed via the Beltrami regularization based on the color image manifold theory. Comparative experiments on a variety of test images demonstrate the power and the applicability of the proposed method.

Identification and integrity verification of sealed containers for security applications can be obtained by employing noninvasive portable optical systems. We present a portable laser range imaging system capable of identifying welds, a byproduct of a container’s physical sealing, with micrometer accuracy. It is based on the assumption that each weld has a unique three-dimensional (3-D) structure which cannot be copied or forged. We process the 3-D surface to generate a normalized depth map which is invariant to mechanical alignment errors and that is used to build compact signatures representing the weld. A weld is identified by performing cross correlations of its signature against a set of known signatures. The system has been tested on realistic datasets, containing hundreds of welds, yielding no false positives or false negatives and thus showing the robustness of the system and the validity of the chosen signature.

High-efficiency video coding (HEVC) is a new coding standard that adopts the quadtree splitting structure based on coding tree units instead of macroblocks, and can support more coding modes and more partitions. Although it can improve compression efficiency, the flexible quadtree block partition and mode selection result in high computational complexity in real-time applications. We propose a low-complexity video coding algorithm for HEVC by utilizing statistical correlation and spatiotemporal correlation, which consists of an early determination of SKIP mode (EDSM) method and an early termination of reference frame selection (ETRFS) method. Since there is a strong correlation for the rate distortion (RD) cost for the SKIP mode between adjacent frames, EDSM detects the SKIP mode according to the threshold derived from the former training frame. Meanwhile, ETRFS terminates the process of reference frame selection using the motion vector and reference frame information from neighboring blocks to skip unnecessary candidate frames. Experimental results demonstrate that the proposed method can achieve about 45.01% complexity reduction on average with a 1.11% BD-rate increase and 0.04 BD-PSNR decrease for random access. The complexity reduction, BD-rate increase, and BD-PSNR decrease for low delay are 46.16%, 0.99%, and 0.03, respectively.

Motivated by the remarkable successes of graph-based transduction (GT) and sparse representation (SR), we present a classifier named sparse graph-based classifier (SGC) for image classification. In SGC, SR is leveraged to measure the correlation (similarity) of every two samples and a graph is constructed for encoding these correlations. Then the Laplacian eigenmapping is adopted for deriving the graph Laplacian of the graph. Finally, SGC can be obtained by plugging the graph Laplacian into the conventional GT framework. In the image classification procedure, SGC utilizes the correlations which are encoded in the learned graph Laplacian, to infer the labels of unlabeled images. SGC inherits the merits of both GT and SR. Compared to SR, SGC improves the robustness and the discriminating power of GT. Compared to GT, SGC sufficiently exploits the whole data. Therefore, it alleviates the undercomplete dictionary issue suffered by SR. Four popular image databases are employed for evaluation. The results demonstrate that SGC can achieve a promising performance in comparison with the state-of-the-art classifiers, particularly in the small training sample size case and the noisy sample case.

Tampering related to document forgeries is often accomplished by copy-pasting or add-printing. These tampering methods introduce character distortion mutation in documents. We present a method of exposing document forgeries using distortion mutation of geometric parameters. We estimate distortion parameters, which consist of translation and rotation distortions, through image matching for each character. Detection of tampered characters with distortion mutation occurs based on a distortion probability, which is calculated from character distortion parameters. The introduction of a visualized probability map describes the degree of distortion mutation for a full page. The proposed method exposes the forgeries based on individual characters and applies to English and Chinese document examinations. Experimental results demonstrate the effectiveness of our method on low JPEG compression quality and low resolution.

Sparse representation has been successfully applied to pattern recognition problems in recent years. The most common way for producing sparse coding is to use the l₁-norm regularization. However, the l₁-norm regularization only favors sparsity and does not consider locality. It may select quite different bases for similar samples to favor sparsity, which is disadvantageous to classification. Besides, solving the l₁-minimization problem is time consuming, which limits its applications in large-scale problems. We propose an improved algorithm for sparse coding and dictionary learning. This algorithm takes both sparsity and locality into consideration. It selects part of the dictionary columns that are close to the input sample for coding and imposes locality constraint on these selected dictionary columns to obtain discriminative coding for classification. Because an analytic solution of the coding is derived by only using part of the dictionary columns, the proposed algorithm is much faster than the l₁-based algorithms for classification. Besides, we also derive an analytic solution for updating the dictionary in the training process. Experiments conducted on five face databases show that the proposed algorithm has better performance than the competing algorithms in terms of accuracy and efficiency.

In order to segment a noised image, a method is proposed based on the rough set and orthogonal polynomial density model, in which the nonparametric mixture model can accurately fit the image gray distribution and the rough set can deal with the inaccuracy and uncertainty problems. First, the nonparametric mixture density model is constructed based on the upper and lower approximations of the rough set which can address the problem of over-relying on the prior presumption. Second, the nonparametric expectation-maximization is used to estimate the mixture model parameters. Finally, image pixels are classified according to Bayesian criterion. Experiments on different datasets show that our method is effective in solving the problem of model mismatch, restraining the noise, and preserving the boundary for the noised image segmentation.

We present a complexity reduction scheme for the depth map intraprediction of three-dimensional high-efficiency video coding (3-D-HEVC). The 3-D-HEVC introduces a new set of specific tools for depth map coding, inserting additional complexity to intraprediction, which results in new challenges in terms of complexity reduction. Therefore, we present the DMMFast (depth modeling modes fast prediction), a scheme composed of two new algorithms: the simplified edge detector (SED) and the gradient-based mode one filter (GMOF). The SED anticipates the blocks that are likely to be better predicted by the traditional intramodes, avoiding the evaluation of DMMs. The GMOF applies a gradient-based filter in the borders of the block and predicts the best positions to evaluate the DMM 1. Software evaluations showed that DMMFast is capable of achieving a time saving of 11.9% on depth map intraprediction, considering the random access mode, without affecting the quality of the synthesized views. Considering the all intraconfigurations, the proposed scheme is capable of achieving, on average, a time saving of 35% considering the whole encoder. Subjective quality assessment was also performed, showing that the proposed technique inserts minimal quality losses in the final encoded video.

We propose an interactive object segmentation method which learns feature-specific segmentation parameters based on a single image. The first step is to design discriminative features for each pixel, which integrate four kinds of cues, i.e, the color Gaussian mixture model (GMM), the graph learning-based attribute, the texture GMM, and the geodesic distance. Then we formulate the segmentation problem as a conditional random field model in terms of fusing multiple features. While an image-specific parameter setting is practical in interactive segmentation, the efficiency of learning process highly depends on the type of user interaction and the designed features. We propose a feature-specific parameter learning strategy to learn model parameters, in which an offline training stage is not required and parameters are computed according to some sparsely labeled pixels on the basis of a single image. Extensive experiments show that the proposed segmentation model performs well for segmenting images with a weak boundary, texture, or cluttered background. Comparative experiment results demonstrate that our method can achieve both qualitative and quantitative improvements over other state-of-the-art interactive segmentation methods.

Representation-based classification methods, such as sparse representation-based classification, have been a breakthrough for face recognition recently. Under such a philosophy, we develop a framework that fuses virtual synthesized training samples as bases and fuzzy discriminant sparse residuals measurement in face classification. More specifically, the preprocessing of the proposed algorithm aims to alleviate sampling uncertainty by introducing extra training samples and then the features are extracted over the above dictionary using a hierarchical multiscale local binary patterns scheme. The second stage tries to approximate testing samples by a subset of training samples, and the introduced sparse coding process with weak l₁-constraint has superior competitiveness in that the accuracy has been improved while the complexity has fallen. The third stage again determines a new weighted sum for the remaining informative samples. Hence, fuzzy sparse similarity grades are designed by the new weighted value, which can be merged into the typical discriminant analysis criterion. Experimental results from various benchmark face databases have demonstrated the effectiveness of our algorithm.

We formulate a simple human-pose tracking theory from monocular video based on the fundamental relationship between changes in pose and image motion vectors. We investigate the natural embedding of the low-dimensional body pose space into a high-dimensional space of body configurations that behaves locally in a linear manner. The embedded manifold facilitates the decomposition of the image motion vectors into basis motion vector fields of the tangent space to the manifold. This approach benefits from the style invariance of image motion flow vectors, and experiments to validate the fundamental theory show reasonable accuracy (within 4.9 deg of the ground truth).

Facial expressions are an important demonstration of humanity’s humors and emotions. Algorithms capable of recognizing facial expressions and associating them with emotions were developed and employed to compare the expressions that different cultural groups use to show their emotions. Static pictures of predominantly occidental and oriental subjects from public datasets were used to train machine learning algorithms, whereas local binary patterns, histogram of oriented gradients (HOGs), and Gabor filters were employed to describe the facial expressions for six different basic emotions. The most consistent combination, formed by the association of HOG filter and support vector machines, was then used to classify the other cultural group: there was a strong drop in accuracy, meaning that the subtle differences of facial expressions of each culture affected the classifier performance. Finally, a classifier was trained with images from both occidental and oriental subjects and its accuracy was higher on multicultural data, evidencing the need of a multicultural training set to build an efficient classifier.

Halftoning is a crucial part of image reproduction in print. First-order frequency modulated (FM) halftones, in which the single dots are stochastically distributed, are widely used in printing technologies, such as inkjet, that are able to stably print isolated dispersed dots. Printers, such as laser printers, that utilize electrophotographic technology are not able to stably print the isolated dots and, therefore, use clustered-dot halftones. Periodic clustered-dot, i.e., amplitude modulated halftones are commonly used in this type of printer, but they suffer from an undesired periodic interference pattern called moiré. An alternative solution is to use second-order FM halftones in which the clustered dots are stochastically distributed. The iterative halftoning techniques that usually result in well-formed halftones operate on the whole input image and require extensive computations and thus, are very slow when the input image is large. We introduce a method to generate image-independent threshold matrices for first- and second-order FM halftoning. The first-order threshold matrix generates well-formed halftone patterns and the second-order FM threshold matrix can be adjusted to produce clustered dots of different sizes, shapes, and alignment. Using predetermined and image-independent threshold matrices makes the proposed halftoning method a point-by-point process and thereby very fast.

This paper proposes multimodal biometric-based personal identification. Palmprint and fingerprint modalities are utilized in the proposed model due to the availability of more distinctive features and the ease of capture. Principle lines are the main discriminative features in the palmprint, whereas orientations of the ridge and valley structures are the main features by which to identify the fingerprint. To extract these features, the use of the Radon transform is proposed in this work. However, the Radon transform is sensitive to the orientation. In order to make the model rotation invariant and insensitive to noise, a normalization process is generally used. Here, a logarithm-based normalization process has been utilized in the proposed model. A Euclidean-based matching process that is invariant to the rotation has been used. The proposed model is applicable to low resolution images and is invariant to rotation, insensitive to noise, and has less computational complexity as compared to other models.

Obtaining high-quality depth maps and disparity maps with the use of a stereo camera is a challenging task for some kinds of objects. The quality of these maps can be improved by taking advantage of a larger number of cameras. The research on the usage of a set of five cameras to obtain disparity maps is presented. The set consists of a central camera and four side cameras. An algorithm for making disparity maps called multiple similar areas (MSA) is introduced. The algorithm was specially designed for the set of five cameras. Experiments were performed with the MSA algorithm and the stereo matching algorithm based on the sum of sum of squared differences (sum of SSD, SSSD) measure. Moreover, the following measures were included in the experiments: sum of absolute differences (SAD), zero-mean SAD (ZSAD), zero-mean SSD (ZSSD), locally scaled SAD (LSAD), locally scaled SSD (LSSD), normalized cross correlation (NCC), and zero-mean NCC (ZNCC). Algorithms presented were applied to images of plants. Making depth maps of plants is difficult because parts of leaves are similar to each other. The potential usability of the described algorithms is especially high in agricultural applications such as robotic fruit harvesting.

A framework for crowd modeling using a combination of multiple kernel learning (MKL)-based fast head detection and shape-aware matching is proposed. First, the MKL technique is used to train a classifier for head detection using a combination of the histogram of oriented gradient and local binary patterns feature sets. Further, the head detection process is accelerated by implementing the classification procedure only at those spatial locations in the image where the gradient points overlap with moving objects. Such moving objects are determined using an adaptive background subtraction technique. Finally, the crowd is modeled as a deformable shape through connected boundary points (head detection) and matched with the subsequent detection from the next frame in a shape-aware manner. Experimental results obtained from crowded videos show that the proposed framework, while being characterized by a low computation load, performs better than other state-of-art techniques and results in reliable crowd modeling.

An end-to-end license plate recognition system is proposed. It is composed of preprocessing, detection, segmentation, and character recognition to find and recognize plates from camera-based still images. The system utilizes connected component (CC) properties to quickly extract the license plate region. A two-stage CC filtering is utilized to address both shape and spatial relationship information to produce high precision and to recall values for detection. Floating peak and valleys of projection profiles are used to cut the license plates into individual characters. A turning function-based method is proposed to quickly and accurately recognize each character. It is further accelerated using curvature histogram-based support vector machine. The INFTY dataset is used to train the recognition system, and MediaLab license plate dataset is used for testing. The proposed system achieved 89.45% F-measure for detection and 87.33% accuracy for overall recognition rate which is comparable to current state-of-the-art systems.

Slow feature analysis (SFA) extracts slowly varying features out of the input data and has been successfully applied on pattern recognition. However, SFA heavily relies on the constructed time series when SFA is applied on databases that neither have obvious temporal structure nor have label information. Traditional SFA constructs time series based on k-nearest neighborhood (k-NN) criterion. Specifically, the time series set constructed by k-NN criterion is likely to include noisy time series or lose suitable time series because the parameter k is difficult to determine. To overcome these problems, a method called adaptive unsupervised slow feature analysis (AUSFA) is proposed. First, AUSFA designs an adaptive criterion to generate time series for characterizing submanifold. The constructed time series have two properties: (1) two points of time series lie on the same submanifold and (2) the submanifold of the time series is smooth. Second, AUSFA seeks projections that simultaneously minimize the slowness scatter and maximize the fastness scatter to extract slow discriminant features. Extensive experimental results on three benchmark face databases demonstrate the effectiveness of our proposed method.

With the advent of progressive format display and broadcast technologies, video deinterlacing has become an important video-processing technique. Numerous approaches exist in the literature to accomplish deinterlacing. While most earlier methods were simple linear filtering-based approaches, the emergence of faster computing technologies and even dedicated video-processing hardware in display units has allowed higher quality but also more computationally intense deinterlacing algorithms to become practical. Most modern approaches analyze motion and content in video to select different deinterlacing methods for various spatiotemporal regions. We introduce a family of deinterlacers that employs spectral residue to choose between and weight control grid interpolation based spatial and temporal deinterlacing methods. The proposed approaches perform better than the prior state-of-the-art based on peak signal-to-noise ratio, other visual quality metrics, and simple perception-based subjective evaluations conducted by human viewers. We further study the advantages of using soft and hard decision thresholds on the visual performance.

In the practice of line detection applications based on Radon transform (RT), noise and clutter decrease the sharpness of the RT local peaks, which correspond to linear edges in the original image. We suggest a new approach to line detection on gray-scale images using the so-called weighted Radon transform (WRT). The proposed WRT-based approach exploits gradient direction information in such a way that only the gradient component perpendicular to the line direction is integrated to form a local peak corresponding to the line. Theoretical and experimental studies have shown the effectiveness of the suggested WRT-based line detection method.

Compressive sensing (CS) theory is popular and has led to the development of new imaging methods in many fields. Specifically, CS theory can reduce the sampling rate of a high-accuracy star tracker and maintain the quality of star images. However, the imaging system of a star tracker contains certain properties that cannot correspond to conventional measurement architectures. Thus, a compressive imaging system that is suitable for the star trackers used in space missions is proposed. First, the CS framework is briefly introduced and the feasibility of typical measurement architectures is analyzed. Second, an imaging method with a low data rate and based on deterministic phase modulation (DPM) is presented to form a weight block circulant matrix (WBCM) of deterministic measurement. The advantage of the proposed imaging method is determined by comparing it with the typical measurement architectures. Third, numerical simulations are conducted to investigate the performance of the proposed method. Results show that the reconstruction quality of the DPM architecture is close to that of random phase modulation architecture. Meanwhile, the WBCM measurement is better than the conventional block circulant matrix measurement. Furthermore, the statistical result significantly ensures accurate star centroid estimation and confirms the feasibility of CS application to star tracker imaging at low data rates.

Sparse representation has been applied to an online subspace learning-based tracking problem. To handle partial occlusion effectively, some researchers introduce l₁ regularization to principal component analysis (PCA) reconstruction. However, in these traditional tracking methods, the representation of each object observation is often viewed as an individual task so the inter-relationship between PCA basis vectors is ignored. We propose a new online visual tracking algorithm with multitask sparse prototypes, which combines multitask sparse learning with PCA-based subspace representation. We first extend a visual tracking algorithm with sparse prototypes in multitask learning framework to mine inter-relations between subtasks. Then, to avoid the problem that enforcing all subtasks to share the same structure may result in degraded tracking results, we impose group sparse constraints on the coefficients of PCA basis vectors and element-wise sparse constraints on the error coefficients, respectively. Finally, we show that the proposed optimization problem can be effectively solved using the accelerated proximal gradient method with the fast convergence. Experimental results compared with the state-of-the-art tracking methods demonstrate that the proposed algorithm achieves favorable performance when the object undergoes partial occlusion, motion blur, and illumination changes.

Single-image super-resolution (SR) is one of the most important and challenging issues in image processing. To produce a high-resolution image from a low-resolution image, one of the conventional approaches is to leverage regularization to overcome the limitations caused by the modeling. However, conventional regularizers such as total variation always neglect the high-level structures in the data. To overcome the drawback, we propose to explore the underlying information for the images with structured edges by using directional total variation. An alternating direction method of a multiplier-based algorithm is presented to effectively solve the resulting optimization problem. Computer simulations on several texture images such as a leaf image have been used to demonstrate the effectiveness and improvement of the proposed method on SR reconstruction, both qualitatively and quantitatively. Furthermore, the effect of parameter selection is also discussed for the proposed method.

Optical coherence tomography (OCT) is a high-resolution imaging technology capable of capturing a three-dimensional (3-D) representation of fingertip skin. The papillary junction—a junction layer of skin containing the same topographical features as the surface fingerprint—is contained within this representation. The top edge of the papillary junction contains the topographical information pertinent to the internal fingerprint. Extracting the internal fingerprint from OCT fingertip scans has been shown to be possible. Currently, acquiring the internal fingerprint involves manually defining the region containing it. This manner of definition is inefficient. Perfect knowledge of the location of the papillary junction is hypothesized as achievable. This research details and tests a k-means clustering approach for papillary junction detection. All tested metrics are of a standard comparable to the measured human error. The technique presented in this research is highly successful in detection of the location of the papillary junction. Furthermore, high-quality internal fingerprints are acquired using the coordinates obtained.

As a breast-imaging technique, digital breast tomosynthesis has great potential to improve the diagnosis of early breast cancer over mammography. Ray-tracing-based reconstruction algorithms, such as ray-tracing back projection, maximum-likelihood expectation maximization (MLEM), ordered-subset MLEM (OS-MLEM), and simultaneous algebraic reconstruction technique (SART), have been developed as reconstruction methods for different breast tomosynthesis systems. This paper provides a comparative study to investigate these algorithms by computer simulation and phantom study. Experimental results suggested that, among the four investigated reconstruction algorithms, OS-MLEM and SART performed better in interplane artifact removal with a fast speed convergence.

Image completion techniques are required to fill in missing regions in digital images. We propose a method that focuses on the complex structure of the foreground object. We observe that if we match the complete deformable contour of the sample object and obtain the missing contour (i.e., point set positions), the structure of the damaged object can be reconstructed. We further find that the reconstructed contour provides sufficient information to complete the image, which generates contour guidance maps as soft constraints. With the contour guidance maps, we then fill in the incomplete image parts by combining textures of both the damaged image and the sample object using optimization techniques. Experimental results demonstrate that our method is a potentially superior method for filling target regions of incomplete foreground objects with complex structure.

It is important to understand how humans view images and how their behavior is affected by changes in the properties of the viewed images and the task they are given, particularly the task of scoring the image quality (IQ). This is a complex behavior that holds great importance for the field of image-quality research. This work builds upon 4 years of research work spanning three databases studying image-viewing behavior. Using eye-tracking equipment, it was possible to collect information on human viewing behavior of different kinds of stimuli and under different experimental settings. This work performs a cross-analysis on the results from all these databases using state-of-the-art similarity measures. The results strongly show that asking the viewers to score the IQ significantly changes their viewing behavior. Also muting the color saturation seems to affect the saliency of the images. However, a change in IQ was not consistently found to modify visual attention deployment, neither under free looking nor during scoring. These results are helpful in gaining a better understanding of image viewing behavior under different conditions. They also have important implications on work that collects subjective image-quality scores from human observers.

A new multimodal expression-invariant face recognition method is proposed by extracting features of rigid and semirigid regions of the face which are less affected by facial expressions. Dual-tree complex wavelet transform is applied in one decomposition level to extract the desired feature from range and intensity images by transforming the regions into eight subimages, consisting of six band-pass subimages to represent face details and two low-pass subimages to represent face approximates. The support vector machine has been used to classify both feature fusion and score fusion modes. To test the algorithm, BU-3DFE and FRGC v2.0 datasets have been selected. The BU-3DFE dataset was tested by low intensity versus high intensity and high intensity versus low intensity strategies using all expressions in both training and testing stages in different levels. Findings include the best rank-1 identification rate of 99.8% and verification rate of 100% at a 0.1% false acceptance rate. The FRGC v2.0 was tested by the neutral versus non-neutral strategy, which applies images without expression in training and with expression in the testing stage, thereby achieving the best rank-1 identification rate of 93.5% and verification rate of 97.4% at a 0.1% false acceptance rate.

We propose a color-image segmentation algorithm by unsupervised classification of pixels. The originality of the proposed approach consists in iteratively identifying pixel classes by taking into account both the pixel color distributions in several color spaces and their spatial arrangement in the image. In order to overcome the difficult problem of the color space choice, the algorithm selects the color space that is well suited to construct the class at each iteration step. The selection criterion is based on connectedness and color homogeneity measures of pixel subsets. In order to tune the sensitivity of segmentation, we introduce a hierarchical criterion that allows us to segment images with different numbers of regions as human observers do. Experiments carried out on the well-known Berkeley segmentation dataset show that this multicolor space approach succeeds in constructing classes that effectively correspond to regions in the image.

Surgical simulation is a major application in computer graphics and virtual reality, and most of the existing work indicates that interactive real-time cutting simulation of soft tissue is a fundamental but challenging research problem in virtual surgery simulation systems. More specifically, it is difficult to achieve a fast enough graphic update rate (at least 30 Hz) on commodity PC hardware by utilizing traditional triangle-based rendering algorithms. In recent years, point-based rendering (PBR) has been shown to offer the potential to outperform the traditional triangle-based rendering in speed when it is applied to highly complex soft tissue cutting models. Nevertheless, the PBR algorithms are still limited in visual quality due to inherent contrast distortion. We propose an adaptive image contrast enhancement algorithm as a postprocessing module for PBR, providing high visual rendering quality as well as acceptable rendering efficiency. Our approach is based on a perceptible image quality technique with automatic parameter selection, resulting in a visual quality comparable to existing conventional PBR algorithms. Experimental results show that our adaptive image contrast enhancement algorithm produces encouraging results both visually and numerically compared to representative algorithms, and experiments conducted on the latest hardware demonstrate that the proposed PBR framework with the postprocessing module is superior to the conventional PBR algorithm and that the proposed contrast enhancement algorithm can be utilized in (or compatible with) various variants of the conventional PBR algorithm.

Foreground segmentation in video frames is quite valuable for object and activity recognition, while the existing approaches often demand training data or initial annotation, which is expensive and inconvenient. We propose an automatic and unsupervised method of foreground segmentation given an unlabeled and short video. The pixel-level optical flow and binary mask features are converted into the normal probabilistic superpixels, therefore, they are adaptable to build the superpixel-level conditional random field which aims to label the foreground and background. We exploit the fact that the appearance and motion features of the moving object are temporally and spatially coherent in general, to construct an object-like pool and background-like pool via the previous segmented results. The continuously updated pools can be regarded as the “prior” knowledge of the current frame to provide a reliable way to learn the features of the object. Experimental results demonstrate that our approach exceeds the current methods, both qualitatively and quantitatively.

The feature matching step plays a critical role during the copy-move forgery detection procedure. However, when several highly similar features simultaneously exist in the feature space, current feature matching methods will miss a considerable number of genuine matching feature pairs. To this end, we propose a clustering-based method to collect qualified matching features for the feature point-based methods. The proposed method can collect far more genuine matching features than existing methods do, and thus significantly improve the detection performance, especially for multiple pasting cases. Experimental results confirm the efficacy of the proposed method.

The latest high-efficiency video coding (HEVC) standard adopts a recursive quad-tree block structure with the coding-unit (CU) size variable depending on video content. It substantially improves the coding efficiency and also dramatically increases complexity. Therefore, a fast CU size selection algorithm based on hierarchical quad-tree correlations (HQTCs) is proposed. First, for each coding tree unit, the partition information at each depth is recorded in a table that reflects the appearance of the quad-tree structure. Then, by using two techniques called top omitting and bottom pruning, the size of the current CU can be determined according to the subtree distributions of adjacent CUs instead of traversing all the depths. Additionally, a gray level co-occurrence matrix-based method is also introduced to further speedup the searching process. Experimental results show that the proposed algorithm can achieve on average a 26% computational time reduction under all configurations with a negligible BD-rate (Bjøntegaard Delta bitrate) increase of 0.47% compared with the original encoding scheme in HEVC test model HM13.0.

We present a surface completeness algorithm that is capable of denoising, removing outliers, and filling in missing patches on point clouds or surfaces. The main advantages of the proposed algorithm include its ability to remove outliers while preserving the details and ability to recover large missing patches. Additionally, our algorithm is a global method, whereby linear programming results are applied to a global optimization problem. This is advantageous because it yields a sparse solution and avoids local minima. Experiments further demonstrate the effectiveness of our algorithm through applications to point clouds where noise, outliers, and large missing patches exist.

Differences of data distributions widely exist among datasets, i.e., domains. For many pattern recognition, nature language processing, and content-based analysis systems, a decrease in performance caused by the domain differences between the training and testing datasets is still a notable problem. We propose a domain adaptation method called domain adaptive boosting (DAB). It is based on the AdaBoost approach with extensions to cover the domain differences between the source and target domains. Two main stages are contained in this approach: source-domain clustering and source-domain sample selection. By iteratively adding the selected training samples from the source domain, the discrimination model is able to achieve better domain adaptation performance based on a small validation set. The DAB algorithm is suitable for the domains with large scale samples and easy to extend for multisource adaptation. We implement this method on three computer vision systems: the skin detection model in single images, the video concept detection model, and the object classification model. In the experiments, we compare the performances of several commonly used methods and the proposed DAB. Under most situations, the DAB is superior.