Face Recognition Method Based on the Adaptive Fuzzy Weighted Sub-Pattern SVD
Yuan Wei · 2017
Adaptive fuzzy weighted sub-pattern SVD algorithm (AFW-SpSVD) is presented in this paper to solve the problem that singular value decomposition (SVD) descriptor on the whole facial image can not provide enough information for face recognition. Unlike traditional SVD based on the whole image pattern, the key of the AFW-SpSVD operates directly on its sub-patterns partitioned from an original whole facial image and separately extracts singular value features from them, so the rich information can be obtained for recognizing human face. The way of establishing feature vectors based on local singular value decomposition is proposed. In the recognition step, the feature vectors of input facial image are set up, and then the membership degrees of these features to each facial sample image are computed, respectively. Based on the obtained membership degrees, the contribution of each part, which is used as weight for fusion each sub-pattern, can be adaptively computed applying specific formula. Finally, the AFW-SpSVD endows them to a classification task in order to enhance the robustness to variations due to expression and illumination. Experiments on three standard face databases show that the proposed method is competitive.