Stratification-based keyframe cliques for removal of near-duplicates in video search results

  • Authors:

  • Xiangang Cheng;Liang-Tien Chia

  • Affiliations:

  • Nanyang Technological University, Singapore, Singapore;Nanyang Technological University, Singapore, Singapore

  • Venue:
  • Proceedings of the international conference on Multimedia information retrieval
  • Year:
  • 2010

Quantified Score

Hi-index 0.00

Visualization

Abstract

The current volume of videos available for distribution or viewing on the internet is increasing exponentially, there is an urgent need for designing effective and efficient video management systems. However, due to the tremendous amounts of video data, it is highly likely that any large scale video systems will provide query results with near-duplicates videos in the return list of videos. In this paper, we introduce our method of identification and removal of near-duplicates in video search results via matching strata of keyframes. To be exact, we detect the near duplicate keyframes in each video separately. Then we partition these keyframes into summarized groups by our quasi-clique based partition. Experiments on the Trecvid dataset confirmed our initial view that a significant number of keyframes from videos in the Trecvid corpus are near-duplicates.Based on the summarized clique representation of each video, we tested our algorithm for detection of near-duplicates in the web videos. Results show that our proposed method greatly speeds up the retrieval process, while maintaining a high detection accuracy.