Image Completion For Complex Objects Based On Exemplar Matching And Filling

Image completion is a very important topic in computer vision, which is widely used in image denoising, photo restoration, the removal of video text and specified objects. The traditional methods mainly based on partial differential equations and exemplar. However, previous methods cannot well solve the problem of image completion on complex objects as objects contain complex structure and texture information. Therefore, image completion on complex objects is a worth studying problem.Accordingly, we combine the data-driven approach and propose an image completion method based on exemplar matching and filling. We first get the contour of object through some user interaction. Then our system will search the labeled database to find the best matched sample and build the patch sets. Finally, the defected object will be restored by a contour rebuilding and region filling method with the patch sets.In order to solve the problem of lack of information, we get the patch sets by sampling the best matched candidate. In addition, we propose an exemplar filling method with image separation and avoid mixed problems in the filling process. In experiments, our method can effectively solve the problem of image completion on complex objects. Moreover, our method can provide various candidate results with different specified contour.