Research On Blind Video Restoration Approach For Spatially-varying Motion Blur

The research on blind video restoration approaches is a challenging topic. It has great significance for improving video quality and restoring video information, so it has received more and more attention recently. However, many existing video restoration approaches assume that the PSFs(Point Spread Function) are spatially-invariant. When they are applied to process the video with spatially-varying PSFs, the performance degrades seriously. Hence, this thesis proposed a blind video restoration approach for spatially-varying motion blur. The main works are as follows:① Based on the survey and analysis of blind video restoration technology development status and developing trends, this thesis proposed a blind video restoration approach for spatially-varying motion blur. It used the prior identification from blind image restoration technology and tried to estimate the PSFs and restore the video frame separately. First, the spatially-varying PSFs of the previous frame are estimated by taking full advantage of the rich information available from two consecutive frames. Then, the video frame is deblurred using the estimated PSFs.② In the stage of spatially-varying PSFs estimation, a weighted method was proposed to refine the initial PSFs by utilizing the local similarity existing in the rigid object, so the more accurate PSFs can be obtained.③ In the stage of video restoration, a patch strategy was proposed to restore the spatially-varying motion blur video by combining the total variation(TV) regularization model and the patch aggregation model. The strategy can not only process the spatially-varying blur, but also can obtain the sharper edge and the fewer ringing artifacts along the strong edges.④ The proposed blind video restoration approach for spatially-varying motion blur was tested on real video, and compared with some state-of-the-art works. The more superior performance of the approach was verified.The proposed approach can remove the blur caused by the rigid object which moves along a straight line in the random direction. Compared with the exiting works, the proposed approach has a great of advantages in terms of subjective evaluation.