Fast and high-quality video inpainting by efficiently patch searching and synthesizing

The rapid development of computer technology leads to an explosive growth in video resources in the internet. Efficient video transmission is appealing for quickly deliver and facilely video quality preservation. Video inpainting is the key solution to the quality problem when data loss occurs during the transmission. The technology aims to restore the missing areas with appropriate data from the rest of the video, which is still challenging due to the high requirement on the inpainted result that can keep visual consistency with existing video content. In this paper, we propose a novel video inpainting method, which relies on the optical flow calculation and local pixel-based matching to achieve quick recovery of the missing areas in videos. Especially, we calculate the optical flow characteristics of video and use them to fast locate and search out the best-match patch. The qualitative and quantitative evaluations in DAVIS and YouTube-VOS datasets demonstrate that the proposed method yields competitive inpainted results and outperforms other pixel-based approaches.