Video summarization by spatial-temporal graph optimization

In this paper we present a novel approach for video summarization based on graph optimization. Our approach emphasizes both a comprehensive visual-temporal content coverage and visual coherence of the video summary. The approach has three stages. First, the source video is segmented into video shots, and a candidate shot set is selected from the video shots according to some video features. Second, a dissimilarity function is defined between the video shots to describe their spatial-temporal relation, and the candidate video shot set is modelled into a directional graph. Third, we outline a dynamic programming algorithm and use it to search the longest path in the graph as the final video skimming. A static video summary is generated at the same time. Experimental results show encouraging promises of our approach for video summarization.