In this paper we consider the problem of the pose variance in face recognition. In practice, the face images we get usually have different poses; however the number of training sample is limited, so the sparse representation based classification (SRC) for face recognition cannot work well. For the pur