An Effective Exemplar-based Image Inpainting Method

Image inpainting refers to filling in the missing region of an image using the information from surrounding known region. Criminisi has proposed an exemplar-based image inpainting algorithm, which can propagate structure information and texture information into target region simultaneously. Although this method has such advantages as simplicity and easy implementation, it has some flaws of perceptual quality and computational efficiency. So in this paper, we present an improved method for exemplar-based image inpainting. By considering the isophote curvature as a new ingredient to compute patch priority and matching cost function, this method performs well when the target region contains linear structure with significant changes in curvature. To solve "staircase effect" caused by reusing the same source patch, we do not use the source patch having been used once any more. Experimental results show the superiority of our method in terms of both inpainting quality and computational efficiency.