[Show abstract][Hide abstract] ABSTRACT: In this paper, we propose a new framework for compressive video sensing (CVS) that exploits the inherent spatial and temporal redundancies of a video sequence, effectively. The proposed method splits the video sequence into the key and non-key frames followed by dividing each frame into the small non-overlapping blocks of equal sizes. At the decoder side, the key frames are reconstructed using adaptively learned sparsifying (ALS) basis via $\ell_0$ minimization, in order to exploit the spatial redundancy. Also, three well-known dictionary learning algorithms are investigated in our method. For recovery of the non-key frames, a prediction of the current frame is initialized, by using the previous reconstructed frame, in order to exploit the temporal redundancy. The prediction is employed in a proper optimization problem to recover the current non-key frame. To compare our experimental results with the results of some other methods, we employ pick signal to noise ratio (PSNR) and structural similarity (SSIM) index as the quality assessor. The numerical results show the adequacy of our proposed method in CVS.
Telecommunications (IST), 2014 7th International Symposium on, Tehran, Iran; 09/2014
[Show abstract][Hide abstract] ABSTRACT: In this paper, we study transmission of a memoryless Laplacian source over three types of channels: additive white Laplacian noise (AWLN), additive white Gaussian noise (AWGN), and slow flat-fading Rayleigh channels under both bandwidth compression and bandwidth expansion. For this purpose, we analyze two well-known hybrid digital–analog (HDA) joint source–channel coding schemes for bandwidth compression and one for bandwidth expansion. Then we obtain achievable (absolute-error) distortion regions of the HDA schemes for the matched signal-to-noise ratio (SNR) case as well as the mismatched SNR scenario. Using numerical examples, it is shown that these schemes can achieve a distortion very close to the provided lower bound (for the AWLN channel) and to the optimum performance theoretically attainable bound (for AWGN and Rayleigh fading channels) on mean-absolute error distortion under matched SNR conditions. In addition, a non-linear analog coding scheme is analyzed, and its performance is compared to the HDA schemes for bandwidth compression under both matched and mismatched SNR scenarios. The results show that the HDA schemes outperform the non-linear analog coding over the whole CSNR region.
IEEE Transactions on Communications 07/2014; · 1.98 Impact Factor
[Show abstract][Hide abstract] ABSTRACT: In this paper, first we present an improved method for conventional block-based compressed sensing (BCS) image recovery algorithm called BCS-SPL that deploys smoothed pro-jected Landweber (SPL) iterations for image recovery. In our proposed method a median filter is applied instead of Wiener filter, specifically in low measurement rates. Also, we employ a strict thresholding criterion as an alternative to the universal threshold criterion. We refer to call our proposed method as BCS-ImSPL. Also, we investigate how the BCS-ImSPL can be improved to a faster recovery algorithm, by considering two accelerated strategies, Beck and Teboulle's fast iterative shrinkage thresholding algorithm (FISTA) and Bioucas-Dias and Figueiredo's two-step iterative shrinkage thresholding (TwIST) algorithm. To compare our experimental results with the other methods, we employ the pick signal to noise ratio (PSNR) and the structural similarity (SSIM) index as the quality assessors. Our vast experiments show good performance of the accelerated BCS-ImSPL method for recovery of images in terms of execution time and image quality.
[Show abstract][Hide abstract] ABSTRACT: Due to the optimal sparse representation of objects with edges by the multiscale and directional Curvelet Transform, its application have been increasingly interested over the past years. In this paper, we investigate how the block-based compressed sensing (BCS) can be improved to an efficient recovery algorithm, by employing the iterative Curvelet thresholding (ICT). Also, we consider two accelerated iterative shrinkage thresholding (IST) methods, including the following: 1) Beck and Teboulle's fast iterative shrinkage thresholding algorithm (FISTA); 2) Bioucas-Dias and Figueiredo's two-step iterative shrinkage thresholding (TwIST) algorithm, to increase the execution speed of the proposed methods rather than simple ICT. To compare our experimental results with the results of some other methods, we employ pick signal to noise ratio (PSNR) and structural similarity (SSIM) index as the quality assessor. Numerical results show good performance of the new proposed BCS using accelerated ICT methods, in terms of these two quality assessments.
[Show abstract][Hide abstract] ABSTRACT: Automated human identification is a significant issue in real and virtual societies. Iris is a suitable choice for meeting this goal. In this paper, we present an iris recognition system that uses images acquired in both near-infrared and visible lights. These two types of images reveal different textural information of the iris tissue. We demonstrated the necessity to process both VL and NIR images to recognize irides. The proposed system exploits two feature extraction algorithms: one is based on 1D log-Gabor wavelet which gives a detailed representation of the iris region and the other is based on 1D Haar wavelet which represents a coarse model of iris. The Haar wavelet algorithm is proposed in this paper. It makes smaller iris templates than the 1D log-Gabor approach and yet achieves an appropriate recognition rate. We performed the fusion at the match score level and examined the performance of the system in both verification and identification modes. UTIRIS database was used to evaluate the method. The results were compared with other approaches and proved to have better recognition accuracy, while no image enhancement technique is utilized prior to the feature extraction stage. Furthermore, we demonstrated that fusion can compensate the lack of input image information, which can be beneficial in reducing the computation complexity and handling non-cooperative iris images.
Machine Vision and Applications 05/2014; 25(4):881-899. · 1.44 Impact Factor
[Show abstract][Hide abstract] ABSTRACT: In our previous work, “robust transmission of scalable video stream using modified LT codes”, an LT code with unequal packet protection property was proposed. It was seen that applying the proposed code to any importance-sorted input data, could increase the probability of early decoding of the most important parts when enough number of encoded symbols is available at the decoder’s side. In this work, the performance of the proposed method is assessed in general case for a wide range of loss rate, even when there are not enough encoded symbols at the decoder’s side. Also in this work the degree distribution of input nodes is investigated in more detail. It is illustrated that sorting input nodes in encoding graph, as what we have done in our work, has superior advantage in comparison with unequal input node selection method that is used in traditional rateless code with unequal error protection property.
[Show abstract][Hide abstract] ABSTRACT: Human tracking is an interesting topic in computer vision domain. In this paper, a human detection and tracking algorithm based on new features combination in one camera system is proposed. In detection part, first, mixture of Gaussian background subtraction method is used to find moving regions, then histogram of oriented gradient (HOG) feature of these regions are extracted. At the end, SVM classifier is used to distinguish human from non-human according to their HOG features. In tracking part, first, color, cellular local binary pattern (Cell-LBP) and HOG features of humans are extracted, then their next positions are estimated using particle filter framework. Color, Cell-LBP and HOG features are used to model humans. Color is an effective feature in dealing with object deformation and partial occlusion but has some restriction in cases where background or objects have same color. Cell-LBP is an improved texture descriptor that is robust against partial occlusion, this feature compensates color's restriction. HOG is a shape descriptor that can separate humans from background and is robust against illumination changes. Combination of these three features improves tracking result despite challenges like partial occlusion, object's deformation and illumination changes. Experimental results show advantage of the proposed algorithm.
Machine Vision and Image Processing (MVIP); 09/2013
[Show abstract][Hide abstract] ABSTRACT: In this paper we propose an efficient method for behavior recognition and identification of anomalous behavior in video surveillance data. This approach consists of two phases of training and testing. In the training phase, first, we use background subtraction method to extract the moving pixels. Then optical flow vectors are extracted for moving pixels. We propose behavior features of each pixel as the average all optical flow vectors in the pixel over several frames in video data. Next, we use spectral clustering to classify behaviors wherein pixels that have similar behavior features are clustered together. Then we obtain a behavior model for each cluster using the normal distribution of the samples. Once the behavior models are obtained, in the testing phase, we use these models to detect anomalous behavior in a test video of the same scene. Experimental results on video surveillance sequences show the effectiveness and speed of proposed method.
2013 8th Iranian Conference on Machine Vision and Image Processing (MVIP); 09/2013
[Show abstract][Hide abstract] ABSTRACT: This paper presents a novel and uniform framework for face recognition. This framework is based on a combination of Gabor wavelets, direct linear discriminant analysis (DLDA) and support vector machine (SVM). First, feature vectors are extracted from raw face images using Gabor wavelets. These Gabor-based features are robust against local distortions caused by the variance of illumination, expression and pose. Next, the extracted feature vectors are projected to a low-dimensional subspace using DLDA technique. The Gabor-based DLDA feature vectors are then applied to SVM classifier. A new kernel function for SVM called hyperhemispherically normalized polynomial (HNP) is also proposed in this paper and its validity on the improvement of classification accuracy is theoretically proved and experimentally tested for face recognition. The proposed algorithm was evaluated using the FERET database. Experimental results show that the proposed face recognition system outperforms other related approaches in terms of recognition rate.
[Show abstract][Hide abstract] ABSTRACT: Human tracking is one of the most important topics in surveillance systems. Increment of system's ability to detect and track humans in both indoor and outdoor crowded environments leads to a safer environment. In this paper color and shape information are fused based on particle filter framework to track humans. Histogram of oriented gradient (HOG) is a shape descriptor that is used as a feature to detect humans using support vector machine (SVM) classifier. The first step of human detection is mixture of Gaussian method that is used to find moving regions of the scene, then HOG feature of these regions is extracted and finally SVM is used to distinguish human from non-human. This algorithm leads to a fewer computational complexity against traditional method of human detection that used sliding windows to detect humans. Human motion is non Linear and non-Gaussian so a particle filter framework is used to track human. Color and HOG histograms are used to model humans. Occlusion is one of the most important tracking challenges. According to increment of surveillance requirements, three-camera system is used to handle occlusion. Experimental results show the effectiveness of the proposed algorithm.
[Show abstract][Hide abstract] ABSTRACT: Compressed sensing is a new theory that samples a signal below the Nyquist rate. While Gaussian and Bernoulli random measurements perform quite well on the average, structured matrices such as Toeplitz are mostly used in practice due to their simplicity. However, the signal compression performance may not be acceptable. In this paper, we propose to optimize the Toeplitz matrices to improve its compression performance to recover sparse signals. We establish the optimization on minimizing the coherence of the measurement matrix by an intelligent optimization method called Particle Swarm Optimization. Our simulation results show that the optimized Toeplitz matrix outperforms the non-optimized one in reconstructing sparse signals in terms of quality and sampling rate.
Communication and Information Theory (IWCIT), 2013 Iran Workshop on; 01/2013
[Show abstract][Hide abstract] ABSTRACT: In this paper an effective three view multiple human tracking method based on color and texture information fusion is proposed. Since human motion is usually non-linear and non-Gaussian, a particle filter framework is used to estimate human position. Human model is jointly represented by weighted color and cellular LBP (cellular local binary pattern) histograms. Weighted color histogram is robust to scale invariant and partial occlusion but has a main limitation when object's color and background's color are similar; so using these two complement features improve tracking results. This method is robust against illumination changes and occlusions. A three-camera network is used to handle occlusion. Tracking process has done separately for each camera, when occlusion is detected in one view. Tracking results of two other views are used to handle occlusion. Experimental results demonstrate that the proposed method improves performance of human tracking.
[Show abstract][Hide abstract] ABSTRACT: In this paper, we introduce pedestrian detection using combination of low level features like CNN, HOG and Haar with high level features. Two kinds of high level features were used in this paper. One is related to the probability of existence of human's face, which obtained from combination of skin color and possible location and area for the human's face. The other is related to the probability of existence of human's anti-body which obtained by curvature checking of vertical edges, situation of them relative to each other and location of them in the detection window. Several different structures were studied and their results were compared on a diagram. Also the average execution times of them were gathered in a table. At first, we show that appending the high level features to every low level feature improves the performance of detection very much and then, with proper arrangement of several features, it is possible to improve the performance of detection further without increasing the execution time. For evaluation of the proposed algorithm, INRIA database and a video sequence were used.