Video^M: Multi-video Synopsis

Conventional video representation methods focus predominantly on a single video, aiming at reducing the space-time redundancy as much as possible, while this paper describes a novel approach to simultaneously presenting dynamics of multiple videos, aiming at a less intrusive viewing experience. Given a main video and multiple supplementary videos, the proposed approach automatically constructs a synthesized multi-video synopsis, called Video^M, by integrating the supplementary videos into the most suitable space-time portions within the main video. We formulate the problem of Video^M as a maximum a posterior (MAP) problem which maximizes the desired properties related to less intrusive viewing experience, i.e., informativeness, consistency, visual naturalness, and stability. This problem is solved by the Viterbi beam search algorithm to optimally find the suitable integration between the main video and supplementary videos.