Video Deflickering Using Multi - Frame Optimization

Chao Li, Zhihua Chen, Bin Sheng, Ping Li, Gaoqi He

Abstract: In this paper, we propose an approach of removing flickering artifacts in a video. The video is obtained via applying the image-based processing methods to an original non-flickering video. In traditional video deflickering methods, we always reconstruct flickering frames with nonflickering frames, which fail to keep video spatial consistency and are always designed to address certain video flickering artifact under specific condition. On the contrary, we propose a general multiple frames based video deflickering approach, where we take both temporal and spatial coherence into account. Instead of reconstructing a flickering frame only from its last frame, we warp multiple corresponding frames to reconstruct the flickering frame, so the warp inaccuracy in the reconstruction process can be reduced. By taking the advantage of video fidelity, temporal coherence and spatial coherence, we formulate video deflickeing objective as a least-squares energy. A non-flickering output video can be obtained via solving the constructed energy formulation with the least angle regression. Results of visual quality, objective measurement and user study demonstrate the efficiency of our proposed multiple frames based video deflickering approach.