Kernel-based two-dimensional maximum scatter-difference projection discriminant analysis and face recognition

Traditional kernel methods are only applied to one-dimensional data. The paper develops a kernel-based two-dimensional maximum scatter difference projection analysis method, where the kernel methods can directly be applied into original two-dimensional image matrices which need not be transformed into one-dimensiona vectors. It is able to extract more effective nonlinear feature and improve the correct recognition rates. Whatpsilas more, it also offers a unified framework for kernel-based two-dimensional projection discriminant analysis. Finally, extensive experiments performed on AR face database verify the effectiveness of the proposed method.