Enhanced Exemplar Based Inpainting Using Patch Ratio

Abstract

Image inpainting is the art of modyng part of image or video that cannot detectable by ordinary observer. There are two dierent classes of algorithms for inpainting. Texture synthesis, and geometric partial dierential equations(PDEs). We think that we don't have to make new color in image so we only focus on nd the best matching pixels for each unknown pixels. In this paper, we suggest a new method based on exemplar based inpainting. The method is suggested by Criminisi et al. [1]. But its method takes a long time, but also cannot give a good result for some cases. First, we analyze deciency of Criminisi algorithm and then we oer a new algorithm. Our algorithm uses a modied calculation for priority and then uses patch ratio for nding the best candidates. For faster computing, we use window and number of nonzero elements of candidates. Finally, among the candidates we concern about gradients of the template.