Abstract: The occlusion dictionary of sparse representation based classification algorithm is a highdimensional dictionary without redundancy.To solve this problem, a method to construct occlusion dictionary is proposed in this paper.First, the input image is divided into small blocks and occlusion base images at all levels are attained.Then each unique occlusion base images is expanded into column vectors in dictionary order.Finally, a low-dimensional occlusion dictionary with redundancy is constructed by putting these column vectors together.The experimental results show that the proposed method not only obviously improves the occluded face recognition rate, but also saves sparse decomposition time and improves running efficiency of sparse representation based classification algorithm by reducing the blocking partition level of image.