Variational Video Mask Generation

Video mask, which delimits the ROI (Region Of Interest) of frame, is frequently-used in video editing. Direct interpolation often needs the time-consuming manual modification. In this paper, we present a method of mask generation, which formulates the mask generation as a variation problem, and makes the mask field subject to the dual constraints from the video and keyframes. The differential solution can significantly maintain the spatiotemporal coherence.