Image Completion Based on Partial Region

Abstract

The traditional image completion algorithms based on patches, are to iteratively search the most similar patches in the known region of the whole image to fill the target region. Generally, it is difficult to use these algorithms in practice, due to the big computation. To overcome this problem, in this paper we propose a novel image completion algorithm based on partial regions. Firstly we adapt a randomized correspondence algorithm to search the pixels in the known region that are similar with the pixels in the border of the target region. Then the regions that include the main structures and Texture can be determined by these pixels, and thus we can limit the search space. Secondly we optimize the method of computing filling priorities that is based confidence factor and edge information. Consequently, the correctness of the order of the repairing patches and the success of structures propagation have been enhanced. Finally we improve the method of calculating the most similar patches. A variety of experiments show that, in general, our method yields better results and is 5-10 times faster than the state-of-the-art methods.