Multi-Video Summarization Using Complex Graph Clustering and Mining

Jian Shao, Dongming Jiang, Mengru Wang, Hong Chen, Lu Yao

Multi-video summarization is a great theoretical and technical challenge due to the wider diversity of topics in multi-video than singlevideo as well as the multi-modality nature of multi-video over multidocument. In this paper, we propose an approach to analyze both visual and textual features across a set of videos and to create a so-called circular storyboard composed of topic-representative keyframes and keywords. We formulate the generation of circular storyboard as a problem of complex graph clustering and mining, in which each separated shot from visual data and each extracted keyword from speech transcripts are first structured into a complex graph and grouped into clusters; hidden topics in the representative keyframes and keywords are then mined from clustered complex graph while at the same time maximizing the coverage of the summary over the original video set. We also design experiments to evaluate the effectiveness of our approach and the proposed approach shows a better performance than two other storyboard baselines.