Abstract

We present video co-summarization, a novel perspectiveto video summarization that exploits visual co-occurrenceacross multiple videos. Motivated by the observation thatimportant visual concepts tend to appear repeatedly acrossvideos of the same topic, we propose to summarize a videoby finding shots that co-occur most frequently across videoscollected using a topic keyword. The main technical chal-lenge is dealing with the sparsity of co-occurring patterns,out of hundreds to possibly thousands of irrelevant shots invideos being considered. To deal with this challenge, we de-veloped a Maximal Biclique Finding (MBF) algorithm thatis optimized to find sparsely co-occurring patterns, discard-ing less co-occurring patterns even if they are dominant in one video. Our algorithm is parallelizable with closed-form updates, thus can easily scale up to handle a large numFber of videos simultaneously. We demonstrate the effectiveiness of our approach on motion capture and self-compiled fYouTube datasets. Our results suggest that summaries genoerated by visual co-occurrence tend to match more closely pwith human generated summaries, when compared to sevDeral popular unsupervised techniques. t a