Distorted image restoration via non-rigid registration and lucky-region fusion approach

Image distortion occurs in various environments, such as water waves, atmosphere turbulence, etc. Original image is distorted and needs to be restored for further applications. In this paper, we propose a framework to resolve image distortion, which combines non-rigid image registration technique with lucky-region fusion approach to produce an image of high quality. To improve registration performance, residual complexity measure is employed to substitute traditional measures such as SSD, MI. Also, an additional step of refining image after each registration is taken to get rid of sharp edges. We test our algorithm on real datasets and it shows the efficiency of our method.