Abstract: In order to alleviate the influence of illumination variations on face recognition,in this paper,we proposed an effective discriminative sparse representation face recognition method．Firstly,we decomposed test image by using nonsubsampled contourlet transformation (NSCT),then used coefficients of each subbands to calculate the contribution of each subbands and fusion the subbands information according to the contribution of each classified image block．Finally,we obtained the discriminative character of the test image,applied it in the face recognition．The experiment on AR database shows that the proposed algorithm is more robust in illumination and expression variations in face recognition．

Key words: Face recognition,Sparse representation,Discriminative,Nonsubsampled contourlet transformation