An Improved Kernel Fisher Discriminant Analysis for Face Recognition

A weighted kernel maximum scatter difference discriminate criterion is developed for extraction of nonlinear feature. The proposed method not only extracts nonlinear feature for faces effectively, but also reconstructs between-class and within-class scatter matrix by weighted schemes. So it can modify the kernel maximum scatter difference discriminate criterion function. Considering this method sensitive to the change of illumination, a pretreatment strategy that can reduce image gradation is used. Finally experiments performed on ORL and Yale face database verify the effectiveness of the proposed method.