A Supervised Locality Preserving Projections Based Local Matching Algorithm for Face Recognition

In this paper, a novel local matching algorithm based on supervised locality preserving projections (LM-SLPP) is proposed for human face recognition. Unlike the holistic face recognition methods which operates directly on the whole face images and obtains a global face features, the proposed LM-SLPP operates on sub-patterns partitioned from the original whole face image and separately extracts corresponding local sub-features from them. In our method, the input face images are firstly divided into several sub-images. Then, the supervised locality preserving projections is applied on each sub-image set for feature extraction. At last, the nearest neighbor classifier combined with major voting is utilized to classify the new face images. The efficiency of the proposed algorithm is demonstrated by experiments on Yale and YaleB face databases. Experimental results show that LM-SLPP outperforms other holistic and sub-pattern based methods.