Youjun Xiang

Structured occlusion coding for robust face recognition

Occlusion in face recognition is a common yet challenging problem. While sparse representation based classification (SRC) has been shown promising performance in laboratory conditions (i.e. noiseless or random pixel corrupted), it performs much worse in practical scenarios. In this paper, we consider the practical face recognition problem, where the occlusions are predictable and available for sampling. We propose the structured occlusion coding (SOC) to address occlusion problems. The structured coding here lies in two folds. On one hand, we employ a structured dictionary for recognition. On the other hand, we propose to use the structured sparsity in this formulation. Specifically, SOC simultaneously separates the occlusion and classifies the image. In this way, the problem of recognizing an occluded image is turned into seeking a structured sparse solution on occlusion-appended dictionary. In order to construct a well-performing occlusion dictionary, we propose an occlusion mask estimating technique via locality constrained dictionary (LCD), showing striking improvement in occlusion sample. On a category-specific occlusion dictionary, we replace l1 norm sparsity with the structured sparsity which is shown more robust, further enhancing the robustness of our approach. Moreover, SOC achieves significant improvement in handling large occlusion in real world. Extensive experiments are conducted on public data sets to validate the superiority of the proposed algorithm.

Robust Sparse Signal Recovery in the Presence of the S αS Noise.

In this letter, robust sparse signal recovery is considered in the presence of the symmetric α-stable distributed noise. An M-estimate type model is constructed by approximating the location score function of the noise. A reweighed iterative hard thresholding algorithm is proposed to recover the sparse signal. The basis functions for the approximation and the recovery performance of the proposed algorithm are discussed. Simulations are given to demonstrate the validity of our results.

A novel low-rank model for MRI using the redundant wavelet tight frame

Abstract The low-rank matrix reconstruction has been attracted significant interest in compressed sensing magnetic resonance imaging (CS-MRI). To the end of computability, rank is often modeled by nuclear norm. The singular value thresholding (SVT) algorithm is taken as a solver of this model, usually. However, this model with the solver may be insufficient to obtain a high quality magnetic resonance (MR) image at high speed. Still inspired by the low-rank matrix reconstruction idea, we proposes a novel low-rank model with a new scheme of the weight selection to reconstruct the MR image under the redundant wavelet tight frame. A fast and accurate solver is given for the proposed model. Further, a new scheme is presented to accelerate the proposed solver. Numerical experiments demonstrate that the proposed solver and its accelerated version can converge stably. The proposed method is faster than some existing methods with the comparable quality.

A Novel Model for Compressed Sensing MRI via Smoothed e1-Norm Regularization

Compressed sensing magnetic resonance imaging (CS-MRI) using ℓ1-norm minimization has been widely and successfully applied. However, ℓ1-norm minimization often leads to bias estimation and the solution is not as accurate as desired. In this paper, we propose a novel model for MR image reconstruction, which takes as a smoothed ℓ1-norm regularization model that is convex, has a unique solution. More specifically, we employ the logarithm function with the parameter in our optimization, and an iteration technique is developed to solve the proposed minimization problem for MR image reconstruction efficiently. The model is simple and effective in the solution procedure. Simulation results on normal brain image demonstrated that the performance of the proposed method was better than some traditional methods.

A Lorentzian IHT for Complex-Valued Sparse Signal Recovery

In this paper, robust complex-valued sparse signal recovery is considered in the presence of impulse noise. A generalized Lorentzian norm is defined for complex-valued signals. A complex Lorentzian iterative hard thresholding algorithm is proposed to realize the signal recovery. Simulations are given to demonstrate the validity of our results.

Candidate of Initial Value in Lloyd Algorithm for Constructing Low-Coherence Matrix

The problem of constructing matrix with lowcoherence is arised in many applications, such as CDMA, compressive sensing (CS), beamforming, etc. Usually the design of low-coherence codebook can be modeled as vector quantization (VQ) problem, and generalized Lloyd algorithm is designed to solve it. Since Lloyd algorithm is a method of local optimization, its performance is influenced by initial value. In this paper, candidate of initial value of Lloyd algorithm is studied. A construction based on finite abelian group which can be used as candidate of initial value will be analyzed. Experimental results prove that using this construction as initial value in Lloyd algorithm can improve the performance significantly.

Efficient locality-constrained occlusion coding for face recognition

Abstract Occlusion is a common yet challenging problem in face recognition. Most of the existing approaches cannot achieve the accuracy of the recognition with high efficiency in the occlusion case. To address this problem, this paper proposes a novel algorithm, called efficient locality-constrained occlusion coding (ELOC), improving the previous sparse error correction with Markov random fields (SEC_MRF) algorithm. The proposed approach estimates and excludes occluded region by locality-constrained linear coding (LLC), which avoids the time-consuming l 1 -minimization and exhaustive subject-by-subject search during the occlusion estimation, and greatly reduces the running time of recognition. Moreover, by simplifying the regularization, the ELOC can be further accelerated. Experimental results on several face databases show that our algorithms significantly improve the previous algorithms in efficiency without losing too much accuracy.

Sparse support recovery using correlation information in the presence of additive noise

The correlation based framework has recently been proposed for sparse support recovery in noiseless case. To solve this framework, the constrained least absolute shrinkage and selection operator (LASSO) was employed. The regularization parameter in the constrained LASSO was found to be a key to the recovery. This paper will discuss the sparse support recoverability via the framework and adjustment of the regularization parameter in noisy case. The main contribution is to provide noise-related conditions to guarantee the sparse support recovery. It is pointed out that the candidates of the regularization parameter taken from the noise-related region can achieve the optimization and the effect of the noise cannot be ignored. When the number of the samples is finite, the sparse support recoverability is further discussed by estimating the recovery probability for the fixed regularization parameter in the region. The asymptotic consistency is obtained in probabilistic sense when the number of the samples tends to infinity. Simulations are given to demonstrate the validity of our results.

Selection of the regularization parameter in the P-LASSO for the noisy covariance model

In this paper, the Positive constrained Least Absolute Shrinkage and Selection Operator (P-LASSO) is studied for sparse support recovery using the correlation information in Compressive sensing (CS). A structural constraint is obtained for selecting the regularization parameter in the case of additive Gaussian noise. Since the measurements are finite in practice, the probability of successful recovering the sparse support is discussed. A lower bound of the probability is derived. Experimental results are provided to illustrate the validity of our main results.

A joint classification approach via sparse representation for face recognition

We consider the problem of automatically recognizing human faces in which sparse representation-based classification (SRC) offers a key. SRC includes two steps: seeking sparest solution and making decision by dictionary classifier (DC). Aiming at improving the performance of face recognition, this paper proposes a joint classification approach based on sparse representation. We initialize dictionary with part of the training samples and train a linear classifier (LC) with the remaining. Thus, the joint classifier (JC), which combines the DC and LC, can decide which subject the query image belongs to. To validate the joint classifier, a residual-based evaluating criterion is established to measure the classification reliability for two classifiers. Experimental results verify that the proposed joint classification strategy significantly improves recognition accuracy at the cost of affordable computational complexity.