Abstract: Kernel Fisher discriminant analysis method is an effective nonlinear discriminant analysis method. Tradition kernel Fisher discriminant analysis uses only single kernel function, which makes it insufficient in face feature extraction,therefore we proposed multiple kernel Fisher discriminant analysis. Weighted projections were obtained through weighted combination several projections obtained by single kernel Fisher discriminant,and then feature extraction and classification were made based on the weighted projections. Experimental results show that weighted multiple kernel Fisher discriminant analysis method is superior to single kernel Fisher discriminant analysis for facial feature extraction and classification.

Key words: Multiple kernel learning,Kernel methods,Facial feature extraction