Face Recognition based on Sub-pattern Sparsity Preserving Projection

In order to solve the problem of pseudo approach in SPP, an unsupervised algorithm named sub-pattern sparsity preserving projection(SpSPP) was proposed in this paper. In the proposed algorithm, face images are firstly divided into smaller sub-images, and sub-images from the same location are collected to compose the sub-pattern set. Then the conventional SPP is applied to each of sub-pattern sets to extract the local features Finally, the sub-pattern features computed by SPP are concatenated to get the holistic features. Based on the fact that different regions of face images share a different similarity relationship and the discrimination information of sparse representation, SpSPP alleviated the effect of pseudo approach through image partition and feature concatenation. The effectiveness of the proposed method was verified on popular face databases (AR and Yale B).