Research Of Face Recognition Algorithm Based On Improved KPCA And LDA

With the rapid development of science and technology, various kinds of information security technology emerge in an endless stream, and researchers always seek for the ultimate goal of the high recognition rate. Biometric technology(such as fingerprint, iris recognition, face recognition, etc.) is one of the most popular information security technology now. Especially for face recognition algorithms based on subspace in face recognition in recent years. However, the face recognition rate can be seriously affected by various external factors, so most researchers consider extracting the most useful information in faces and improving the face recognition rate as significant tasks.Through a lot of literature, paper author learned and analyzed the shortcomings of existing face recognition, and then proposed a improved algorithm KPCA and LDA integration. Firstly, we solved nonlinear problems of the face; then we reduced the sample dimension and solve the problems of "small sample" and the edge data classification;finally we classified the fusion with the use of improved KNN methods and SVM.The research work this paper have done is as follows:(1)To primitive face image albino, low-pass filtering was pretreated to remove interference and noise, at the same time to balance the image power spectrum. Then we extracted facial eight directions feature vectors using dual-tree complex wavelet and Gabor wavelet, as a preparation for the formation of future feature space.(2)In order to solve the over complex computing problem as a result of the non-linear and huge dimension of the sample, kernel principal component analysis algorithm(KPCA) was put forward using the best sample mean estimate vector instead of the original sample vector; to solve the problems of "small sample" and edge data classification, we proposed an improved LDA algorithm. That is, process the between class scatter matrix using specific weighted value processing, and then extract the null space of within class scatter matrix and remove the null space of the between class scatter matrix. As a result, these two methods are combined to propose the KPCA and LDA improved fusion algorithm.(3)To further reduce the dimension of the sample and suppress noise interference effectively, we put forward the method of improved KNN and support vector machine fusion.(4)Performances of various types of face recognition algorithms were tested in the platform of MATLAB 2010 a. Results of amounts of experiments in this article includes that face recognition algorithm we proposed has a significant advantage in recognition rate and the rate.