A Novel Technique to Remove Marked Dynamic Object from Video Based on Reweighted Low Rank Tensor Completion

摘要

This paper focused on the process of missing area reconstruction in a video after removing a unwanted moving object, thus can remove or handle the dynamic occlusion. Dynamic occlusion is the result of one moving object in a 3-D spread tackling another object from view, and also it is a challenging problem in many video processing applications like object tracking, video surveillance, movie production etc. Discarding the moving objects from videos is a very time consuming task, that even a visual effects expert generally clear up with a time-taking manual frame-by-frame rephrasing. This paper proposes a new technique to remove dynamic occlusion in video, even in the case of videos with complex scenes and also compared with the other existing method, the proposed method does not require the exact shape, size and structure of the object. Here, the proposed method effectively taking the advantage of the spatio-temporal correlation in the video to refill the hole created by a cut-out object by utilizing the information available on other part of the video, where the occluded object were viewable.