Kernel-Based Foley-Sammon Discriminant Analysis and Face Recognition

Through implementing two equivalent models of kernel-based Foley-Sammon discriminant analysis （KFSDA） and studying the relationship between the solutions of these two models,a new approach of solving the KFSDA model is presented and proved.Analysis shows that KFSDA retains FSDA’s advantage of distinctly reducing the redundant information among components of the pattern samples,and more importantly,it can extract nonlinear features effectively,thus greatly enhancing the capability of FSDA. Experimental results on ORL face database indicate that the proposed method is valid.