Search

Linear Regression Classification (LRC) is a newly-appeared pattern recognition method, which formulates the recognition problem in terms of class-specific linear regression with sufficient training samples per class. In this paper, we extend LRC via intraclass variant dictionary and SVD to undersampled face recognition where there are very few, or even only one, training sample per class. Intraclass variant dictionary is adopted in undersampled situation to represent the possible variation between the training and testing samples. Three types of methods, quasi-inverse, ridge regularization and Singular Value Decomposition (SVD), are designed to solve low-rank problem of data matrix. Then the whole algorithm, named Extended LRC (ELRC), is presented for face recognition via intraclass variant dictionary and SVD. The experimental results on three well-known face databases show that the proposed ELRC has better generalization ability and is more robust to classification than many state-of-the-art methods in undersampled situation. [ABSTRACT FROM AUTHOR]

In this paper, we propose a stereo matching algorithm based on distance transform to generate high-quality disparity maps with occlusion handling. In general, pixel intensities around object edges are smeared due to mixed values located between the object and its background. This leads to problems when identifying discontinuous disparities. In order to handle these problems, we present an edge control function according to distance transform values. Meanwhile, occluded regions occur, i.e., some portions are visible only in one image. An energy function is designed to detect such regions considering warping, cross check, and luminance difference constraints. Consequently, we replace the disparity in the occluded region with the one chosen from its neighboring disparities in the non-occluded region based on color and spatial correlations. In particular, the occlusion hole is filled according to region types. Experimental results show that the proposed method outperforms conventional stereo matching algorithms with occlusion handling. [ABSTRACT FROM AUTHOR]

Abstract: Dictionary learning has been a hot topic fascinating many researchers in recent years. Most of existing methods have a common character that the sequences of learned dictionaries are simpler and simpler regularly by minimizing some cost function. This paper presents a novel predual dictionary learning (PDL) algorithm that updates dictionary via a simple gradient descent method after each inner minimization step of Predual Proximal Point Algorithm (PPPA), which was recently presented by Malgouyres and Zeng (2009) [F. Malgouyres, T. Zeng, A predual proximal point algorithm solving a non negative basis pursuit denoising model, Int. J. Comput. Vision 83 (3) (2009) 294–311]. We prove that the dictionary update strategy of the proposed method is different from the current ones because the learned dictionaries become more and more complex regularly. The experimental results on both synthetic data and real images consistently demonstrate that the proposed approach can efficiently remove the noise while maintaining high image quality and presents advantages over the classical dictionary learning algorithms MOD and K-SVD. [Copyright &y& Elsevier]

Abstract: In this paper, we propose to use Chambolle’s dual methods to solve Total Variation (TV) inpainting model and (weighted) TV colorization model. The fidelity coefficients in these two models are functions which taking zero in the inpainting region and a positive constant in the other region. Then Chambolle’s dual method can not be directly used to solve these models since the fidelity coefficient will be denominator in the algorithm. In order to overcome this drawback, we propose to approximate these models by adding new variables. Then the approximated problems can be solved by alternating minimization method with Chambolle’s dual method and closed form solutions which is fast and easy to implement. Mathematical results of existence of minimizers are proved for both the original and the approximated problems. Numerical results and comparison with other closely related methods demonstrate that our algorithms are quite efficient. [Copyright &y& Elsevier]