Enhanced adaptive patch based exemplar image inpainting

Abstract

Image inpainting is a process in which missing or deteriorated portions of an image is reconstructed. In the proposed framework, an adaptive patch updating strategy for the basic Criminisi's exemplar-based image inpainting is implemented. Patch matching is the supreme stage of exemplar-based inpainting, but Criminisi uses fixed size patches for the matching process. The proposed method improves the existing approach by adaptively setting the size of patches. A four-level discrete wavelet transform is used to identify whether the host image is a smooth or detailed one. The proposed method could exhibit faster inpainting results in smooth images. For smooth images, it will make large patches, and with these large patches, the algorithm process patch matching and inpainting. Series of experiments conducted for inpainting different images. Compared to existing systems, the outcomes of the proposed framework were favorable in smooth images.