Robust PCA for Face Recognition with Occlusion Using Symmetry Information

Recently, the accuracy and the speed of face recognition have been improved greatly. However, the recognition of faces with occlusions still remains an open problem. A fundamental important problem is the existing methods could not handle large scale occlusion situations well. To address this problem, we propose a novel face recognition methods based on RPCA and facial symmetry, called Symmetry-RPCA. First, we use RPCA to find the sparse errors areas of objective faces; Then, we use the threshold method and k NN method to find the occlusion areas from the sparse errors areas, and we use the mirror parts of face to fix those areas. To compensate for the unnatural characteristic, we do RPCA again with the mirror fixed faces. After the process, the result faces could both keep the traits from the samples and make full use of the information of the original occluded face image. After those, we use CRC for the classification. Experimental results on several popular data sets validate that the proposed methods are of high efficiency and robustness to the recognition of blocked faces, which exceeds existing methods greatly.