Tensor based approach for inpainting of video containing sparse text

Abstract

Videos received from certain sources may contain irrelevant contents which might reduce the amount of information conveyed by it. This paper proposes an effective tensor based video inpainting approach to improve the quality of these videos by removing and replacing unwanted contents with relevant information. The proposed method employs reweighted tensor decomposition technique to identify and discard the inappropriate sparse components of the video data. These sparse components are substituted with proper contents by utilising spatio-temporal consistency through reweighted tensor completion. The replacement is carried out in such a way that the resulting video possesses superior temporal consistency and visual credibility. The proposed method is applied to sparse text removal of videos having dynamic content in various extents and is found that our method outperforms its counterparts.