Face recognition has very large academic and practical values. How enable the machine to have human's intelligence and remember and recognition person just like what human does, is always face recognition research hot spot. The human face particularity enables face recognition technology to become the most potential method of identity recognition, at the same time ,it is difficult to implement face recognition for face concrete shape multiplicity and locates the environment complexity. Face recognition involves the technology are very many, key was the feature extraction and classification. This paper mainly study face extraction and class method ,which concept can be summarized as follows:The preprocessing work includes the image enhancement, the geometry and the gradation normalization and the whiting processing of face images sample.These preprocessing effectively improve image quality, reduce the computation order of complexity and speed up the following core algorithm the convergence rate.In the feature extraction step, this paper regards the independent component analysis(ICA) as the face feature extraction method, simultaneously takes the characteristic choice with the appraisal classifier performance proportionality factor for the basis, thus causes the face feature which extract not only the mutual independence but also classified ability, forms various components mutual statistical independent feature space .In the face recognition, important information not only exists in the picture element in two-order statistical property but also contains in the high-order statistical property. The tradition extraction method based on the principal component analysis(PCA) human only to be able to obtain the face image two-order statistics information and extracted feature is easily influenced by illumination condition. Compares says ICA extracted feature which is sensitive to high-order statistic information in the data and not easily influenced by illumination condition change. Therefore, the ICA method may better identify and reconstruct high-dimensional face image data than PCA. Conventional ICA algorithms exist iteration number of times many and sometimes converge difficultly , therefore, this paper uses FastICA to take ICA the fast algorithm, this algorithm iteration number of times are few, will speed up the convergence of ICA.In the classified distinction, this paper applies the Non- negative Matrix Factorization (NMF)method to face recognition, presents a new subspace classifier method based on NMF and has carried on the comprehensive proof from the algorithm theory and the experiment. This method has inherited the NMF algorithm to the data based on the partial additive expressions characteristic, has to the data because the NMF subspace base has good to the data clamps compels the nature, thus can compact cause the characteristic subspace but to be effective to the data expression. The ORL database in experimental result with the alternative mean comparisons, indicates ICA/NMF unifies the method recognition rate must surpass the traditional method. |