Adaptively weighted subpattern-based sparse preserving projection for face recognition

In this paper, we propose an adaptively weighted subpattern-based sparse preserving projection (Aw-spSPP) algorithm for face recognition. Unlike SPP (Sparse preserving projection) based on a whole image pattern, the proposed AwSpSPP method operates on sub-patterns partitioned from an original whole face image and separately extracts corresponding local sub-features from them. Moreover, the contribution of each sub-pattern can be adaptively computed by sparse weights needless of additional parameter such as neighborhood size used in Aw-spLPP (adaptively weighted subpattern-based locality preserving projection). Experimental results on three bench mark face databases (ORL, YALE and PIE) show that Aw-spSPP can overcome the shortcomings of the existed subpattern-based methods and achieve promising recognition accuracy.