This PDF file contains the editorial “Top Reviewers of 2015” for JEI Vol. 25 Issue 01

This PDF file contains the editorial “2015 Reviewers List” for JEI Vol. 25 Issue 01

The availability of a large number of local invariant feature detectors has rendered the task of evaluating them an important issue in vision research. However, the maximum number of scenes utilized for performance comparison has so far been relatively small. This paper presents an evaluation framework and results based on it utilizing a large number of scenes, providing insights into the performance of local feature detectors under varying JPEG compression ratio, blur, and uniform light changes.

In several image processing applications, discovering regions that have changed in a set of images acquired from a scene at different times and possibly from different viewpoints plays a very important role. Remote sensing, visual surveillance, medical diagnosis, civil infrastructure, and underwater sensing are examples of such applications that operate in dynamic environments. We propose an approach to detect such changes automatically by using image analysis techniques and segmentation based on superpixels in two stages: (1) the tuning stage, which is focused on adjusting the parameters; and (2) the unsupervised stage that is executed in real scenarios without an appropriate ground truth. Unlike most common approaches, which are pixel-based, our approach combines superpixel extraction, hierarchical clustering, and segment matching. Experimental results demonstrate the effectiveness of the proposed approach compared to a remote sensing technique and a background subtraction technique, demonstrating the robustness of our algorithm against illumination variations.

Object detection with intraclass variations is challenging. The existing methods have not achieved the optimal combinations of classifiers and features, especially features learned by convolutional neural networks (CNNs). To solve this problem, we propose an object-detection method based on improved random forest and local image patches represented by CNN features. First, we compute CNN-based patch descriptors for each sample by modified CNNs. Then, the random forest is built whose split functions are defined by patch selector and linear projection learned by linear support vector machine. To improve the classification accuracy, the split functions in each depth of the forest make up a local classifier, and all local classifiers are assembled in a layer-wise manner by a boosting algorithm. The main contributions of our approach are summarized as follows: (1) We propose a new local patch descriptor based on CNN features. (2) We define a patch-based split function which is optimized with maximum class-label purity and minimum classification error over the samples of the node. (3) Each local classifier is assembled by minimizing the global classification error. We evaluate the method on three well-known challenging datasets: TUD pedestrians, INRIA pedestrians, and UIUC cars. The experiments demonstrate that our method achieves state-of-the-art or competitive performance.

Currently, the state-of-the art methods for image denoising are patch-based approaches. Redundant information present in nonlocal regions (patches) of the image is considered for better image modeling, resulting in an improved quality of filtering. In this respect, nonlocal Markov random field (MRF) models are proposed by redefining the energy functions of classical MRF models to adopt a nonlocal approach. With the new energy functions, the pairwise pixel interaction is weighted according to the similarities between the patches corresponding to each pair. Also, a maximum pseudolikelihood estimation of the spatial dependency parameter (β) for these models is presented here. For evaluating this proposal, these models are used as an a priori model in a maximum a posteriori estimation to denoise additive white Gaussian noise in images. Finally, results display a notable improvement in both quantitative and qualitative terms in comparison with the local MRFs.

A low bit rates image compression framework based on adaptive block downsampling and super resolution (SR) was presented. At the encoder side, the downsampling mode and quantization mode of each 16 × 16 macroblock are determined adaptively using the ratio distortion optimization method, then the downsampled macroblocks are compressed by the standard JPEG. At the decoder side, the sparse representation-based SR algorithm is applied to recover full resolution macroblocks from decoded blocks. The experimental results show that the proposed framework outperforms the standard JPEG and the state-of-the-art downsampling-based compression methods in terms of both subjective and objective comparisons. Specifically, the peak signal-to-noise ratio gain of the proposed framework over JPEG reaches up to 2 to 4 dB at low bit rates, and the critical bit rate to JPEG is raised to about 2.3 bits per pixel. Moreover, the proposed framework can be extended to other block-based compression schemes.

The extraction of effective features is extremely important for understanding the intrinsic structure hidden in high-dimensional data. In recent years, sparse representation models have been widely used in feature extraction. A supervised learning method, called sparsity preserving discriminative learning (SPDL), is proposed. SPDL, which attempts to preserve the sparse representation structure of the data and simultaneously maximize the between-class separability, can be regarded as a combiner of manifold learning and sparse representation. More specifically, SPDL first creates a concatenated dictionary by class-wise principal component analysis decompositions and learns the sparse representation structure of each sample under the constructed dictionary using the least squares method. Second, a local between-class separability function is defined to characterize the scatter of the samples in the different submanifolds. Then, SPDL integrates the learned sparse representation information with the local between-class relationship to construct a discriminant function. Finally, the proposed method is transformed into a generalized eigenvalue problem. Extensive experimental results on several popular face databases demonstrate the effectiveness of the proposed approach.

Wireless video multicast has become one of the key technologies in wireless applications. But the main challenge of conventional wireless video multicast, i.e., the cliff effect, remains unsolved. To overcome the cliff effect, a hybrid digital-analog (HDA) video transmission framework based on SoftCast, which transmits the digital bitstream with the quantization residuals, is proposed. With an effective power allocation algorithm and appropriate parameter settings, the residual gains can be maximized; meanwhile, the digital bitstream can assure transmission of a basic video to the multicast receiver group. In the multiple-input multiple-output (MIMO) system, since nonuniform noise interference on different antennas can be regarded as the cliff effect problem, ParCast, which is a variation of SoftCast, is also applied to video transmission to solve it. The HDA scheme with corresponding power allocation algorithms is also applied to improve video performance. Simulations show that the proposed HDA scheme can overcome the cliff effect completely with the transmission of residuals. What is more, it outperforms the compared WSVC scheme by more than 2 dB when transmitting under the same bandwidth, and it can further improve performance by nearly 8 dB in MIMO when compared with the ParCast scheme.

Acquired digital images are often corrupted by a lack of camera focus, faulty illumination, or missing data. An algorithm is presented for fusion of multiple corrupted images of a scene using the lifting wavelet transform. The method employs adaptive fusion arithmetic based on matrix completion and self-adaptive regional variance estimation. Characteristics of the wavelet coefficients are used to adaptively select fusion rules. Robust principal component analysis is applied to low-frequency image components, and regional variance estimation is applied to high-frequency components. Experiments reveal that the method is effective for multifocus, visible-light, and infrared image fusion. Compared with traditional algorithms, the new algorithm not only increases the amount of preserved information and clarity but also improves robustness.

Visual attention models (VAMs) predict the location of image or video regions that are most likely to attract human attention. Although saliency detection is well explored for two-dimensional (2-D) image and video content, there have been only a few attempts made to design three-dimensional (3-D) saliency prediction models. Newly proposed 3-D VAMs have to be validated over large-scale video saliency prediction datasets, which also contain results of eye-tracking information. There are several publicly available eye-tracking datasets for 2-D image and video content. In the case of 3-D, however, there is still a need for large-scale video saliency datasets for the research community for validating different 3-D VAMs. We introduce a large-scale dataset containing eye-tracking data collected from 61 stereoscopic 3-D videos (and also 2-D versions of those), and 24 subjects participated in a free-viewing test. We evaluate the performance of the existing saliency detection methods over the proposed dataset. In addition, we created an online benchmark for validating the performance of the existing 2-D and 3-D VAMs and facilitating the addition of new VAMs to the benchmark. Our benchmark currently contains 50 different VAMs.

We describe an object classification method based on weighted score-level feature fusion using learned weights. Our method is able to recognize 20 object classes in a customized fruit dataset. Although the fusion of multiple features is commonly used to distinguish variable object classes, the optimal combination of features is not well defined. Moreover, in these methods, most parameters used for feature extraction are not optimized and the contribution of each feature to an individual class is not considered when determining the weight of the feature. Our algorithm relies on optimizing a single feature during feature selection and learning the weight of each feature for an individual class from the training data using a linear support vector machine before the features are linearly combined with the weights at the score level. The optimal single feature is selected using cross-validation. The optimal combination of features is explored and tested experimentally using a customized fruit dataset with 20 object classes and a variety of complex backgrounds. The experiment results show that the proposed feature fusion method outperforms four state-of-the-art fruit classification algorithms and improves the classification accuracy when compared with some state-of-the-art feature fusion methods.

An improved context-based adaptive binary arithmetic coding (CABAC) is presented. The idea for the improvement is to use a more accurate mechanism for estimation of symbol probabilities in the standard CABAC algorithm. The authors’ proposal of such a mechanism is based on the context-tree weighting technique. In the framework of a high-efficiency video coding (HEVC) video encoder, the improved CABAC allows 0.7% to 4.5% bitrate saving compared to the original CABAC algorithm. The application of the proposed algorithm marginally affects the complexity of HEVC video encoder, but the complexity of video decoder increases by 32% to 38%. In order to decrease the complexity of video decoding, a new tool has been proposed for the improved CABAC that enables scaling of the decoder complexity. Experiments show that this tool gives 5% to 7.5% reduction of the decoding time while still maintaining high efficiency in the data compression.

Since the end-user of video-based systems is often a human observer, prediction of user-perceived video quality (PVQ) is an important task for increasing the user satisfaction. Despite the large variety of objective video quality measures (VQMs), their lack of generalizability remains a problem. This is mainly due to the strong dependency between PVQ and video content. Although this problem is well known, few existing VQMs directly account for the influence of video content on PVQ. Recently, we proposed a method to predict PVQ by introducing relevant video content features in the computation of video distortion measures. The method is based on analyzing the level of spatiotemporal activity in the video and using those as parameters of the anthropomorphic video distortion models. We focus on the experimental evaluation of the proposed methodology based on a total of five public databases, four different objective VQMs, and 105 content related indexes. Additionally, relying on the proposed method, we introduce an approach for selecting the levels of video distortions for the purpose of subjective quality assessment studies. Our results suggest that when adequately combined with content related indexes, even very simple distortion measures (e.g., peak signal to noise ratio) are able to achieve high performance, i.e., high correlation between the VQM and the PVQ. In particular, we have found that by incorporating video content features, it is possible to increase the performance of the VQM by up to 20% relative to its noncontent-aware baseline.

We describe a method for detecting heads in order to count people in crowded environments using a single camera. The main difference between this method and traditional ones consists of adapting skeleton graph analysis techniques for distinguishing individuals in crowded environments. First, a graph skeleton is calculated for each selected blob in a scene after having performed motion estimation. Then, the structural property of each blob is explored to detect possible heads in order to estimate the number of people. Each detected head in the skeleton silhouette is identified as being in an independent or partial occlusion state and is updated during a tracking process. Finally, the results of our experiments are presented to demonstrate the robustness of our method.

The conventional blue-noise model that specifies the desired noise characteristics of an ideal halftone has been updated recently, and simulation results showed that the updated model can serve as a better guideline for developing halftone algorithms. At the moment, only a feature-preserving multiscale error diffusion-based algorithm was developed based on the updated noise model. As the algorithm does not support real-time applications, a tone-dependent error diffusion (TDED) algorithm is developed based on the updated noise model. To support the proposed TDED algorithm, we optimize a diffusion filter and a quantizer threshold for each possible input gray level based on the updated noise model, such that the algorithm can adapt its diffusion filter and quantizer according to the input intensity value of a pixel to produce a halftone. Simulation results showed that the proposed TDED algorithm can successfully produce halftones bearing the desired noise characteristics as specified by the updated noise model. As a consequence, it provides better performance than conventional error diffusion-based algorithms in terms of various measures including radially averaged power spectrum density and anisotropy. When processing real images, it can eliminate directional artifacts, regular structure patterns, and unintended sharpening effects in its halftoning outputs.

This paper presents a methodology for tracking multiple skaters in short track speed skating competitions. Nonrigid skaters move at high speed with severe occlusions happening frequently among them. The camera is panned quickly in order to capture the skaters in a large and dynamic scene. To automatically track the skaters and precisely output their trajectories becomes a challenging task in object tracking. We employ the global rink information to compensate camera motion and obtain the global spatial information of skaters, utilize random forest to fuse multiple cues and predict the blob of each skater, and finally apply a silhouette- and edge-based template-matching and blob-evolving method to labelling pixels to a skater. The effectiveness and robustness of the proposed method are verified through thorough experiments.

Evaluation of image quality is important for many image processing systems, such as those used for acquisition, compression, restoration, enhancement, or reproduction. Its measurement is often accompanied by user studies, in which a group of observers rank or rate results of several algorithms. Such user studies, known as subjective image quality assessment experiments, can be very time consuming and do not guarantee conclusive results. This paper is intended to help design an efficient and rigorous quality assessment experiment. We propose a method of limiting the number of scenes that need to be tested, which can significantly reduce the experimental effort and still capture relevant scene-dependent effects. To achieve it, we employ a clustering technique and evaluate it on the basis of compactness and separation criteria. The correlation between the results obtained from a set of images in an initial database and the results received from reduced experiment are analyzed. Finally, we propose a procedure for reducing the initial scenes number. Four different assessment techniques were tested: single stimulus, double stimulus, forced choice, and similarity judgments. We conclude that in most cases, 9 to 12 judgments per evaluated algorithm for a large scene collection is sufficient to reduce the initial set of images.

We present an image fusion algorithm for a visible image and a near-infrared image. The proposed algorithm synthesizes a fused image that includes high-visibility information of both images while reducing artifacts caused by geometric and illumination inconsistencies. In the proposed fusion, the high-visibility area is labeled at each pixel by global optimization based on the local visibility and inconsistency. The local visibility is evaluated using a local contrast. The inconsistency is also locally estimated based on a learning-based approach. The fused luminance is constructed using Poisson image reconstruction that preserves the gradient of the selected high-visibility areas. The proposed fusion framework has various applications, which include denoising, haze removal, and image enhancement. Experimental results show that the proposed method has comparable or even superior performance to existing methods designed for specific applications.

Human action classification based on the adaptive key frame interval (AKFI) feature extraction is presented. Since human movement periods are different, the action intervals that contain the intensive and compact motion information are considered in this work. We specify AKFI by analyzing an amount of motion through time. The key frame is defined to be the local minimum interframe motion, which is computed by using frame differencing between consecutive frames. Once key frames are detected, the features within a segmented period are encoded by adaptive motion history image and key pose history image. The action representation consists of the local orientation histogram of the features during AKFI. The experimental results on Weizmann dataset, KTH dataset, and UT Interaction dataset demonstrate that the features can effectively classify action and can classify irregular cases of walking compared to other well-known algorithms.

Radar coincidence imaging (RCI) is a high-resolution imaging technique without the limitation of relative motion between target and radar. In sparsity-driven RCI, the prior knowledge of imaging model requires to be known accurately. However, the phase error generally exists as a model error, which may cause inaccuracies of the model and defocus the image. The problem is formulated using Bayesian hierarchical prior modeling, and the self-calibration variational message passing (SC-VMP) algorithm is proposed to improve the performance of RCI with phase error. The algorithm determines the phase error as part of the imaging process. The scattering coefficient and phase error are iteratively estimated using VMP and Newton’s method, respectively. Simulation results show that the proposed algorithm can estimate the phase error accurately and improve the imaging quality significantly.

In recent years, many saliency models have achieved good performance by taking the image boundary as the background prior. However, if all boundaries of an image are equally and artificially selected as background, misjudgment may happen when the object touches the boundary. We propose an algorithm called weighted contrast optimization based on discriminative boundary (wCODB). First, a background estimation model is reliably constructed through discriminating each boundary via Hausdorff distance. Second, the background-only weighted contrast is improved by fore-background weighted contrast, which is optimized through weight-adjustable optimization framework. Then to objectively estimate the quality of a saliency map, a simple but effective metric called spatial distribution of saliency map and mean saliency in covered window ratio (MSR) is designed. Finally, in order to further promote the detection result using MSR as the weight, we propose a saliency fusion framework to integrate three other cues—uniqueness, distribution, and coherence from three representative methods into our wCODB model. Extensive experiments on six public datasets demonstrate that our wCODB performs favorably against most of the methods based on boundary, and the integrated result outperforms all state-of-the-art methods.

Video analysis technology has become less expensive and more powerful in terms of storage resources and resolution capacity, promoting progress in a wide range of applications. Video-based human action detection has been used for several tasks in surveillance environments, such as forensic investigation, patient monitoring, medical training, accident prevention, and traffic monitoring, among others. We present a method for action identification based on adaptive training of a multilayer descriptor applied to a single classifier. Cumulative motion shapes (CMSs) are extracted according to the number of frames present in the video. Each CMS is employed as a self-sufficient layer in the training stage but belongs to the same descriptor. A robust classification is achieved through individual responses of classifiers for each layer, and the dominant result is used as a final outcome. Experiments are conducted on five public datasets (Weizmann, KTH, MuHAVi, IXMAS, and URADL) to demonstrate the effectiveness of the method in terms of accuracy in real time.

The description of local surface features is a critical step in surface matching and object recognition. We present a descriptor for three-dimensional shapes based on the bispectrum of spherical harmonics (BSH). First, points in a support region of a feature point are used to construct a local reference frame, and a histogram is formed by accumulating the points falling within each bin in the support region. Second, spherical harmonic coefficients of the histogram and its bispectrum are calculated. Finally, the feature descriptor is obtained via principal component analysis. We tested our BSH descriptor on public datasets and compared its performance with that of several existing methods. The results of our experiments show that the proposed descriptor outperforms other methods under various noise levels and mesh resolutions.

In facial expression recognition tasks, different facial expressions are often confused with each other. Motivated by the fact that a learned metric can significantly improve the accuracy of classification, a facial expression recognition algorithm based on local metric learning is proposed. First, k-nearest neighbors of the given testing sample are determined from the total training data. Second, chunklets are selected from the k-nearest neighbors. Finally, the optimal transformation matrix is computed by maximizing the total variance between different chunklets and minimizing the total variance of instances in the same chunklet. The proposed algorithm can find the suitable distance metric for every testing sample and improve the performance on facial expression recognition. Furthermore, the proposed algorithm can be used for vector-based and matrix-based facial expression recognition. Experimental results demonstrate that the proposed algorithm could achieve higher recognition rates and be more robust than baseline algorithms on the JAFFE, CK, and RaFD databases.

This paper presents a monotonic invariant intensity descriptor (MIID) via spectral embedding and nonsubsampled contourlet transform (NSCT). To make the proposed descriptor discriminative, NSCT is used for the construction of multiple support regions. Specifically, the directed graph and the spectral feature vectors of the signless Laplacian matrix are exploited to construct the MIID. We theoretically demonstrate that the proposed descriptor is able to tackle monotonic illumination changes and many other geometric and photometric transformations. We conduct extensive experiments on the standard Oxford dataset and the complex illumination dataset to demonstrate the superiority of proposed descriptor over the existing state-of-the-art descriptors in dealing with image blur, viewpoint changes, illumination changes, and JPEG compression.

This paper presents a suboptimal quantization and transmission scheme for multiscale block-based compressed sensing images over wireless channels. The proposed method includes two stages: dealing with quantization distortion and transmission errors. First, given the total transmission bit rate, the optimal number of quantization bits is assigned to the sensed measurements in different wavelet sub-bands so that the total quantization distortion is minimized. Second, given the total transmission power, the energy is allocated to different quantization bit layers based on their different error sensitivities. The method of Lagrange multipliers with Karush–Kuhn–Tucker conditions is used to solve both optimization problems, for which the first problem can be solved with relaxation and the second problem can be solved completely. The effectiveness of the scheme is illustrated through simulation results, which have shown up to 10 dB improvement over the method without the rate and power optimization in medium and low signal-to-noise ratio cases.

In past decades, many techniques have been used to improve face recognition performance. The most common and well-studied ways are to use the whole face image to build a subspace based on the reduction of dimensionality. Differing from methods above, we consider face recognition as an image classification problem. The face images of the same person are considered to fall into the same category. Each category and each face image could be both represented by a simple pyramid histogram. Spatial dense scale-invariant feature transform features and bag of features method are used to build categories and face representations. In an effort to make the method more efficient, a linear support vector machine solver, Pegasos, is used for the classification in the kernel space with additive kernels instead of nonlinear SVMs. Our experimental results demonstrate that the proposed method can achieve very high recognition accuracy on the ORL, YALE, and FERET databases.

A mixture model with spatial constraint is proposed for image segmentation. This model assumes that the pixel label priors are similar if the pixels are close in geometry. An energy function is defined on the spatial space for measuring the spatial information. We also derive an energy function on the observed data space from the log-likelihood function of the standard mixture model. We estimate the model parameters by minimizing the combination of the two energy functions, using the gradient descent algorithm. Then we use the parameters to compute the posterior probability. Finally, each pixel can be assigned to a class using the maximum a posterior decision rule. Numerical experiments are presented where the proposed method and other mixture model-based methods are tested on synthetic and real-world images. These experimental results demonstrate that the proposed method achieves competitive performance compared with other spatially constrained mixture model-based methods.

This paper addresses the problem of removing spatially varying blur caused by camera motion with the help of inertial measurements recorded during exposure time. By utilizing a projective motion blur model, the camera motion is viewed as a sequence of projective transformations on the image plane, each of which can be estimated from the corresponding inertial data sample. Unfortunately, measurement noise leads to temporally increasing drift in the estimated motion trajectory and can significantly degrade the quality of recovered images. To address this issue, this paper employs capturing a small sequence of images with different exposure settings along with the recorded inertial data. A special arrangement of exposure settings is designed to anchor the correct position of the camera trajectory, followed by a drift correction step, which makes use of the sharp image structures preserved in one of the captured images. The effectiveness of our approach is demonstrated by conducting comparison experiments on both synthetic images and real images.

High efficiency video coding (HEVC) is the latest video compression standard that provides significant performance improvement on the compression ratio compared to all existing video coding standards. The intra-prediction procedure plays an important role in the HEVC encoder, and it is being achieved by providing up to 35 intra-modes with a larger coding unit requiring a high computational complexity that needs to be alleviated. Toward this end, the paper proposes two fast intra-mode decision algorithms that exploit the features of video sequences. First, an early detection of zero transform and quantified coefficients method is applied to generate threshold values employed for early termination of the intra-decision process and hence accelerates the encoding procedure. Another fast intra-mode decision algorithm is elaborated that relies on a refinement technique. Based on statistical analyses of frequently chosen modes, only a small part of the candidate modes is chosen for intra-prediction process, which reduces the complexity of the intra-encoding procedure. The performance of the proposed algorithms is verified through comparative analysis of encoding time, visual image quality, and compression ratio. Compared to HM 10.0, the encoding time reduction can reach 69% with only a slight degradation of image quality and compression ratio.

The performance of conventional denoising algorithms is usually controlled by one or several parameters whose optimal settings depend on the contents of the processed images and the characteristics of the noises. Among these parameters, noise level is a fundamental parameter that is always assumed to be known by most of the existing denoising algorithms (so-called nonblind denoising algorithms), which largely limits the applicability of these nonblind denoising algorithms in many applications. Moreover, these nonblind algorithms do not always achieve the best denoised images in visual quality even when fed with the actual noise level parameter. To address these shortcomings, in this paper we propose a new quality-aware features-based noise level estimator (NLE), which consists of quality-aware features extraction and optimal noise level parameter prediction. First, considering that image local contrast features convey important structural information that is closely related to image perceptual quality, we utilize the marginal statistics of two local contrast operators, i.e., the gradient magnitude and the Laplacian of Gaussian (LOG), to extract quality-aware features. The proposed quality-aware features have very low computational complexity, making them well suited for time-constrained applications. Then we propose a learning-based framework where the noise level parameter is estimated based on the quality-aware features. Based on the proposed NLE, we develop a blind block matching and three-dimensional filtering (BBM3D) denoising algorithm which is capable of effectively removing additive white Gaussian noise, even coupled with impulse noise. The noise level parameter of the BBM3D algorithm is automatically tuned according to the quality-aware features, guaranteeing the best performance. As such, the classical block matching and three-dimensional algorithm can be transformed into a blind one in an unsupervised manner. Experimental results demonstrate that the performance of the BBM3D algorithm is superior to its nonblind version both in visual qualitative evaluations and quantitative comparisons for various images and noise levels.

We propose joint geodesic depth propagation (JGDP) for two-dimensional-to-three-dimensional video conversion. To save substantial human effort and produce high-quality depth maps, we adopt a key-frame depth propagation strategy based on a geodesic distance. First, we perform block matching between current and reference frames to estimate motion vectors (MVs). Second, we conduct depth compensation from the reference depth map for obtaining the current depth map based on MVs. Finally, we refine the depth map using joint geodesic filtering. Experimental results demonstrate that JGDP achieves an average 2.4-dB gain in peak-signal-to-noise ratio and 0.019 gain in structure similarity over state-of-the-art depth propagation methods.

Realistic visualization is crucial for a more intuitive representation of complex data, medical imaging, simulation, and entertainment systems. In this respect, multiview autostereoscopic displays are a great step toward achieving the complete immersive user experience, although providing high-quality content for these types of displays is still a great challenge. Due to the different characteristics/settings of the cameras in the multiview setup and varying photometric characteristics of the objects in the scene, the same object may have a different appearance in the sequences acquired by the different cameras. Images representing views recorded using different cameras, in practice, have different local noise, color, and sharpness characteristics. View synthesis algorithms introduce artifacts due to errors in disparity estimation/bad occlusion handling or due to an erroneous warping function estimation. If the input multiview images are not of sufficient quality and have mismatching color and sharpness characteristics, these artifacts may become even more disturbing. Accordingly, the main goal of our method is to simultaneously perform multiview image sequence denoising, color correction, and the improvement of sharpness in slightly defocused regions. Results show that the proposed method significantly reduces the amount of the artifacts in multiview video sequences, resulting in a better visual experience.

Visual object tracking is one of the most important components in computer vision. The main challenge for robust tracking is to handle illumination change, appearance modification, occlusion, motion blur, and pose variation. But in surveillance videos, factors such as low resolution, high levels of noise, and uneven illumination further increase the difficulty of tracking. To tackle this problem, an object tracking algorithm based on bit-planes is proposed. First, intensity and local binary pattern features represented by bit-planes are used to build two appearance models, respectively. Second, in the neighborhood of the estimated object location, a region that is most similar to the models is detected as the tracked object in the current frame. In the last step, the appearance models are updated with new tracking results in order to deal with environmental and object changes. Experimental results on several challenging video sequences demonstrate the superior performance of our tracker compared with six state-of-the-art tracking algorithms. Additionally, our tracker is more robust to low resolution, uneven illumination, and noisy video sequences.

In order to ensure privacy constraints in biometrics, new algorithms called template protection schemes have been proposed in the last 10 years in the literature. Most of these algorithms require as input a feature having a fixed size. Texture features can be used within this context for fingerprints as the number of minutiae varies in general for different captures. BioHashing is a two authentication factor algorithm that can be used to ensure privacy while using biometrics. In this work, we compare different recent texture features from the literature within the BioHashing scheme while considering many constraints: efficiency, maximal representation size, and constant size description. Experiments are conducted on three fingerprint databases from the FVC competition. Results permit us to conclude on the texture features to be used within this context.

An adaptive windowed range-constrained Otsu method using local information is proposed for improving the performance of image segmentation. First, the reason why traditional thresholding methods do not perform well in the segmentation of complicated images is analyzed. Therein, the influences of global and local thresholdings on the image segmentation are compared. Second, two methods that can adaptively change the size of the local window according to local information are proposed by us. The characteristics of the proposed methods are analyzed. Thereby, the information on the number of edge pixels in the local window of the binarized variance image is employed to adaptively change the local window size. Finally, the superiority of the proposed method over other methods such as the range-constrained Otsu, the active contour model, the double Otsu, the Bradley’s, and the distance-regularized level set evolution is demonstrated. It is validated by the experiments that the proposed method can keep more details and acquire much more satisfying area overlap measure as compared with the other conventional methods.

A visual tracking framework that provides an object detector and tracker, which focuses on effective and efficient visual tracking in surveillance of real-world intelligent transport system applications, is proposed. The framework casts the tracking task as problems of object detection, feature representation, and classification, which is different from appearance model-matching approaches. Through a feature representation of discriminative sparse coding on local patches called DSCLP, which trains a dictionary on local clustered patches sampled from both positive and negative datasets, the discriminative power and robustness has been improved remarkably, which makes our method more robust to a complex realistic setting with all kinds of degraded image quality. Moreover, by catching objects through one-time background subtraction, along with offline dictionary training, computation time is dramatically reduced, which enables our framework to achieve real-time tracking performance even in a high-definition sequence with heavy traffic. Experiment results show that our work outperforms some state-of-the-art methods in terms of speed, accuracy, and robustness and exhibits increased robustness in a complex real-world scenario with degraded image quality caused by vehicle occlusion, image blur of rain or fog, and change in viewpoint or scale.

A two-stage ear recognition framework is presented where two local descriptors and a sparse representation algorithm are combined. In a first stage, the algorithm proceeds by deducing a subset of the closest training neighbors to the test ear sample. The selection is based on the K-nearest neighbors classifier in the pattern of oriented edge magnitude feature space. In a second phase, the co-occurrence of adjacent local binary pattern features are extracted from the preselected subset and combined to form a dictionary. Afterward, sparse representation classifier is employed on the developed dictionary in order to infer the closest element to the test sample. Thus, by splitting up the ear image into a number of segments and applying the described recognition routine on each of them, the algorithm finalizes by attributing a final class label based on majority voting over the individual labels pointed out by each segment. Experimental results demonstrate the effectiveness as well as the robustness of the proposed scheme over leading state-of-the-art methods. Especially when the ear image is occluded, the proposed algorithm exhibits a great robustness and reaches the recognition performances outlined in the state of the art.

Many nuclear norm minimization (NNM)-based methods have been proposed for hyperspectral image (HSI) mixed denoising due to the low-rank (LR) characteristics of clean HSI. However, the NNM-based methods regularize each eigenvalue equally, which is unsuitable for the denoising problem, where each eigenvalue stands for special physical meaning and should be regularized differently. However, the NNM-based methods only exploit the high spectral correlation, while ignoring the local structure of HSI and resulting in spatial distortions. To address these problems, a total variation (TV)-regularized weighted nuclear norm minimization (TWNNM) method is proposed. To obtain the desired denoising performance, two issues are included. First, to exploit the high spectral correlation, the HSI is restricted to be LR, and different eigenvalues are minimized with different weights based on the WNNM. Second, to preserve the local structure of HSI, the TV regularization is incorporated, and the alternating direction method of multipliers is used to solve the resulting optimization problem. Both simulated and real data experiments demonstrate that the proposed TWNNM approach produces superior denoising results for the mixed noise case in comparison with several state-of-the-art denoising methods.