A patch-based approach to restore videos using additional stills

Abstract

In this paper we introduce a novel technique that restores video frames by using several high quality skills taken of the same static scene. Our method is relied on the robustness of the local feature points. We first select several seed frames by employing a powerful detector/descriptor between video frames and reference photographs. Next, based on this information good local feature points are tracked between adjacent frames. After the geometric distortion between corresponding patches is inferred properly, the higher frequencies are patch-wise transfered to every video frame. Compared with standard unsampling techniques, besides on the important resolution improvement, our approach is able to enhance the overall aspect of the video frames, reducing artifacts such noise and blur and also adjusting pleasantly the color of the contrast parameters. Objective tests on real and synthetic video sequences demonstrate the utility and the benefits of the proposed technique over related methods.