A patch-based image inpainting based on structure consistence

Abstract

Image inpainting scheme is that repairs the damaged region of interest by user specified in an image. For the purpose of patching up automatically, filling in large-sized damaged areas in the image, and lowering down the operation complexity, we design a patch-based inpainting based on structure and texture as our main framework. To keep the structure consistency between the damaged areas and the nondamaged areas, we redefine two factors, confidence and illumination variation, which use to determine the priority of the filling order of target (damaged) regions. With these two defined factors, we propose an algorithm to obtain the best patch in each filling step for the target region. To solve the challenging problem of repairing structure in image completion, we design a novel image completion approach based on automatic salient structure propagation. Experimental results show that the proposed method can give good inpainting and visual quality.