THU-ICRC at rush summarization of TRECVID 2007

  • Authors:

  • Tao Wang;Yue Gao;Jianguo Li;Patricia P. Wang;Xiaofeng Tong;Wei Hu;Yimin Zhang;Jianmin Li

  • Affiliations:

  • Intel China Research Center, Beijing, China;Intel China Research Center, Beijing, China;Intel China Research Center, Beijing, China;Intel China Research Center, Beijing, China;Intel China Research Center, Beijing, China;Intel China Research Center, Beijing, China;Intel China Research Center, Beijing, China;Intel China Research Center, Beijing, China

  • Venue:
  • Proceedings of the international workshop on TRECVID video summarization
  • Year:
  • 2007

Quantified Score

Hi-index 0.00

Visualization

Abstract

In this paper, we describe the THU-ICRC system for the rush summarization task of TRECVID07. Our main objective is to abstract a minimal length rush video by removing useless (or low-quality) and redundant frames and reserving important objects and events by video parsing, cast indexing and important factor analysis. In detail, by video parsing and cast indexing, our approach first constructs story boards to let user know about the main scenes and main actors in the video. Then it detects and removes useless frames, e.g. color bar, near-monochrome/ abrupt/shaking frames, and clap boards etc. Finally, we construct the video skimming by key frame clustering, important factor analysis and repetitive segment detection. Particularly, by the two-stage redundancy removing in both key frame level and video sequence level, we achieve a better performance to shorten the video length. Extensive experiments were carried out on 42 testing videos. Good results demonstrate the effectiveness of the proposed method.