An exemplar-based image inpainting method with search region prior

Abstract

This paper proposes an exemplar-based image inpainting method with adaptive search region. In traditional exemplar-based image inpainting method, it is possible that the number of known pixels in target patch is too small according to the shape of missing region or known pixels of target patch does not contain meaningful features of adjacent known region. In either case, target patch cannot provide enough information to distinguish correct patch in search region that includes wrong candidate patches with different color or texture. Since missing region shape or known pixel values in target patch is not fully adjustable, we propose the method to provide search region with more consistent color or texture to given target patch. The proposed method partitions input image and generate region index map to selectively limit the search region for given target patch. This approach improves the performance of patch searching and provides more correct inpainting result compared to traditional exemplar-based inpainting method.