A curvature constraint Exemplar-based image inpainting

Digital image inpainting belongs to image recovery field, which is to fill the region where information is missing or to replace the missing region with different information. In this article, we mainly study the Exemplar-based method, which was proposed by Criminisi. The Exemplar-based method is to choose appropriate exemplar patches from a known image or the known region of the image waiting to be processed to fill the damaged region. By using Exemplar-based image inpainting method, we maintain the structure information by defining the priority for each unknown pixel and then complete pixels according to each pixel´s priority. In this article, we try to improve the Exemplar-based method aiming at undistinguished priority values and failure in inpainting regions which contain linear structure with significant changes in curvature. Meanwhile, we reasonably optimize the strategy for finding the most similar exemplar from the known region. As a result, our method acquires more reasonable results for some images with both structure and texture missing regions.