Rushes video summarization using audio-visual information and sequence alignment

This paper describes our system and methodologies for the BBC rushes video summarization task of TRECVID 2008. The procedure of the system is composed of three major steps: shot detection, irrelevant and repetitive subshot removal, and final summary generation. First, we segment the original rushes video into subshots according to the difference and accumulative difference of local color histogram between consecutive frames. Second, we recognize the irrelevant subshots, such as subshots of color bar, pure gray frames, and clapper board. We propose a novel video sequence alignment algorithm to detect repetitive subshots. After removing the irrelevant and repetitive subshots, we generate the final summary using the remaining informative and representative subshots. The evaluation from TRECVID 2008 shows that our system can generate good video summaries.