Linear discriminant analysis based on Fisher criterion is one of effective methods for feature extraction

and it was successfully utilized for face recognition. But face image data distribution in practice is highly complex because of illumination

facial expression and pose variations. So it is necessary to extract nonlinear features for face recognition. A novel method called subspace discriminant analysis based on kernel trick is presented in this paper. In the new approach

the kernel trick is used firstly to project the original samples into an implicit space called feature space by nonlinear kernel mapping

then two equivalent models based on generalized Fisher criterion have established by the Theory of Reproducing Kernel in the feature space

and the optimal discriminant vectors are solved finally by using the technique of orthogonal complementary space. The proposed algorithm was tested and evaluated on the ORL face database and the NUST603 face database

which can reach recognition rate shuch as 94% and 99.58%

respectively. The experimental results show that the novel method outperforms both KPCA in[8

9] and Kernel fisherfaces in[ 13] and is comparable with the method in [ 14 ] in terms of correct recognition rate.