Abstract: Because of kernel discriminant analysis(KDA) algorithm only considers c-1 discriminated features, and neglects to capture the boundary structure while computing between-class scatter matrix So an improved algorithm of non-parametric and nonlinear(kerncl) was proposed. It added a weight function during computing the between-class scatter matrix which can overcome the above two disadvantages of KDA. Simulation results show that the recognition performance of the new method is superior to those of the existed methods, and it can avoid using the singular value decomposition theory of the matrixes,so it has some practical value.

Key words: Kernel discriminant analysis, Non-parameter and nonlinear, Face recognition, Feature extraction