Study On Video Completion

Video completion refers to the process of automatically filling in missing pixels with the pixels in the known region. It has application in many areas: video editing, post-processing in the movie-making industry, corrupted internet video repairing and restoration of old movies. Thus, video completion has been one of the hottest areas in computer vision and computer graphics.Firstly, we summarize the existing algorithms on video completion, and introduce various video segmentation algorithms preparing for the proposing of our own algorithm. Secondly, we propose our algorithm: fast video completion based on segmentation. The input video is first segmented into background and moving foreground, then, the background and foreground are repaired respectively by taking advantage of spatial-temporal coherence and periodic movement. As for the background, we complete it along both the time and space axis. Completion along the time axis can make full use of the redundant information in the video while completion along the space axis can involve belief propagation (BP) to propagate known information to unknown regions. Completion of the foreground can be achieved by searching the best match along the time axis, in this way we can preserve motion coherence better and avoid flick and shake at the same time. Finally, we draw a conclusion on our paper and present some future work.