A coarse-to-fine ghost removal scheme for HDR imaging

Ghost removal in high dynamic range imaging is a challenging problem especially when relative camera or object motion exists. To solve the problem, an effective coarse-to-fine deghosting method combining registration and matching based on PatchMatch is proposed. Firstly, the coarse registration scheme based on Scale-Invariant Feature Transform is used to achieve the consistency of image scale space. Secondly, similarity measure of underlayer information is established by Mutual Information to realize fine registration. Thirdly, different from general distance measurement, structural similarity index measurement is employed to build the objective function to search for the best-matched patch in the fusion process. Experimental results demonstrate the algorithm can remove the ghost artifacts effectively. Furthermore, objective evaluations show that the algorithm accuracy has been improved comprehensively. Compared with the existing methods, the proposed algorithm can achieve a convincing result for dynamic senses, especially for large moving objects.