Double layers sparse representation for occluded face recognition

To implement occluded face recognition, we proposed a new method based on sparse representation. First, it partitioned images into four blocks. For each block, unit matrix was added to extend the dictionary and then sparse representation was performed. The coefficients corresponding to the unit matrix were used to estimate the occlusion pixels. Second, the un-occluded part of each block was used to reconstruct each block and the block with the biggest sparsity was used to reconstruct the image again. Occlusion pixels were modified by the residuals. Finally the identification was performed on the un-occluded part. Experiments verify the robustness and effectiveness of our method.