Color and Flow Inpainting for Video Completion

Abstract

We propose a framework for temporally consistent videocompletion. To this end we generalize the exemplar-based inpaintingmethod of Criminisi et al. to video inpainting. Specifically we addresstwo important issues: Firstly, we propose a color and optical flow inpaint-ing to ensure temporal consistency of inpainting even for complex motionof foreground and background. Secondly, rather than requiring the userto hand-label the inpainting region in every single image, we proposea flow-based propagation of user scribbles from the first to subsequentvideo frames which drastically reduces the user input. Experimental com-parisons to state-of-the-art video completion methods demonstrate thebenefits of the proposed approach.

Publication
German Conference on Pattern Recognition, 2014