Image inpainting is the process of reconstructing the missing or damaged contents inside an image in a perceptually plausible way. Although over ten years of research has made the inpainting results suitable for starting to enter real applications, the state-of-the-art methods still suffer from heavy computational complexity and artifacts such as blurring and false edges, especially in large inpainting areas. In this paper we propose a multiresolution approach to image inpainting. In this framework, the image pyramid is first established. Then inpainting is carried out for a chosen low resolution image in the pyramid. Afterwords the information in this repaired low resolution image is propagated through each higher resolution image until the original highest resolution image is inpainted. Compared to the current methods used, the proposed approach holds advantages such as the ease of filling large areas with better results, faster speed, and better conformation with human vision process. Experimental results show that the proposed multiresolution inpainting framework can produce perceptually better results and runs over three times faster than state-of-the-art methods.
展开▼
