基于稳健主成分分析与核稀疏表示的人脸识别

Face Recognition Based on Robust Principal Component Analysis and Kernel Sparse Representation

针对现有人脸识别方法难以有效抑制噪声和误差干扰(如光照、遮挡和表情等)的问题,提出一种基于稳健主成分分析的核稀疏表示分类算法。利用稳健主成分分析将各类训练样本转化为低秩矩阵和误差矩阵之和,并运用这2个矩阵构成稀疏表示的冗余字典。将核稀疏表示问题通过矩阵变换转化为常规的稀疏表示问题,采用正交匹配追踪算法求解该问题得到稀疏表示系数。通过稀疏表示系数计算每个类的重构误差,从而实现人脸识别。实验结果表明,与SRC,ESRC等算法相比,该算法具有较高的人脸识别率,且对噪声和误差干扰有较强的适应能力。

Aiming at the problems that the existing face recognition methods are hard to efficiently overcome the effect of noise and error disturbance（ such as illumination,occlusion,and face expression）. Kernel sparse representation classification based on Robust Principal Component Analysis（ RPCA） is proposed for face recognition. The training sample matrix of each class is decomposed into a low-rank matrix and an error matrix by RPCA algorithm,and the redundant dictionary is constructed by these two matrices. Kernel sparse representation problem is converted to normal sparse representation problem by matrix transformation,and Orthogonal Matching Pursuit（ OMP） technology is used to solve sparse representation problem to obtain sparse representation coefficients. The reconstruction error associated with the each class can be calculated by the sparse coefficients to achieve classification of the test sample. Experimental results show that,compared with Sparse Representation-based Classification（ SRC）,ESRC（ Extended SRC） algorithms,the proposed algorithm has a higher recognition rate and it is robust to noise and error disturbance.