Sparse Representation for Face Verification in Social Insurance System

Face verification system has been applied in the process of social insurance payment to prevent the pension impostors. However, the previous methods of face verification compare the test face with the corresponding training sample in the database, and then simply calculate the similarity of both, ignoring global similarity distribution. In this paper, we put the sparse representation method into the face verification system and propose a new method called SRS. Experiments are performed on six face data sets, such as Yale A, Extended Yale B, AR etc., combing the feature extraction methods of Randomfaces, Eigenfaces, Fisherfaces and Laplacianfaces. Cosine method is used for comparison. Experimental results show that the new method achieves higher accuracy with large numbers of classes.