More training samples can represent more various changes of face image, owing to the variational illumination, facial expressions and poses. However, in real-world applications, due to the limitation of storage capacity and captured image time, the number of face training samples obtained are often limited. In this paper, for the small size problem, we propose to firstly generate various mirror images by the original face images, viewed as its mirror image. Its mirror image is opposite to the original face image with facial details. Then the original sample and its mirror image is respectively used to perform collaborative representation classification method (CSC). Finally, an adaptive weight selection method is proposed to fuse the original face image and its mirror image based on assigning a better weight to the original face image. The results of experiments show that the presented scheme is effective.