Robust scene matching method based on sparse representation and iterative correction

This article presents an efficient scene matching method robust to noise and occlusion. The method combines a coarse matching method with a fine matching method by iterative correction. Both coarse matching method and fine matching method, inspired by sparse representation for face recognition, are resistant to noise and occlusion inherently. In each step of iterative matching, the result of coarse matching is introduced into fine matching as prior knowledge, which gives a rough range about the possible positions. Then, the fine matching finds the most reasonable result based on the rough range given by coarse matching. Finally, the result of fine matching is brought back to coarse matching as post knowledge to correct it. Experiments demonstrate that the robustness to noise and occlusion is improved compared with the matching methods without iterative correction.