The Study On Face Recognition Technology Based On Subspace Methods

Face recognition is one of the personal indentification methods based on biometric. Because of the widely use in security and human-computer interaction, face recognition has been a hottest research area in pattern recognition domain in the last decades. Among all the face recognition methods subspace analysis has become the popular research direction because of its low compute complexity, good description and separation ability. This paper mainly study on subspace analysis, as follows:(1) Three classic subspace mathods PCA, LDA, ICA and their use in face recognition. Their recognition ability is compared by simulation.(2) Different projection vectors has a different influence. A weighted 2DPCA method in which single projection vector recognition rate is used to weighting the projection vectors is proposed. The experiment in ORL face database shows the weighted 2DPCA method has better recognition result.(3) DCT is widely used in the popular image compression methods like JPEG, MPEG. In this thesis we proposed a face recognition method that use DCT and 2DPCA together which greatly reduces the feather dimension and recognition time. Besides 2DPCA are used on the images that have being transformed using block-DCT.The expriment result shows it is workable.(4)Face recognition that bases on two dimension discriminant analysis algorithm does not use the nonlinear feather. We propose a two dimension discriminant analysis algorithm that bases on the kernel method for face recogniton to solve this problem. The effectiveness is proved by expriment.