Abstract: At present,practical face recognition faces the problem of the variation of illumination,expression,pose and the occlusion,disguise.The problem of robust recognizing human faces with varying expression and illumination,as well as occlusion and disguise will be researched in this paper.We cast the test face image as a linear combination of face database and used deformable model representation,Match optimization deformable model for solving combinatorial coefficient,according to the sparse nature of coefficients for classification.Conducting extensive experiments on publicly available databases show that: the average recognition rate is higher than 97%,and the recognition rate is still more than 95% with test image of block 30%.This method is very effective for face recognition and can significantly improve the robustness and stability of disguise or occlusion.