Accelerating video identification by skipping queries with a compact metric cache

  • Authors:

  • Takaaki Aoki;Daisuke Ninomiya;Arnoldo José Müller-Molina;Takeshi Shinohara

  • Affiliations:

  • Department of Artificial Intelligence, Kyushu Institute of Technology, Iizuka, Japan;Department of Artificial Intelligence, Kyushu Institute of Technology, Iizuka, Japan;Department of Artificial Intelligence, Kyushu Institute of Technology, Iizuka, Japan;Department of Artificial Intelligence, Kyushu Institute of Technology, Iizuka, Japan

  • Venue:
  • ICCSA'10 Proceedings of the 2010 international conference on Computational Science and Its Applications - Volume Part IV
  • Year:
  • 2010

Quantified Score

Hi-index 0.00

Visualization

Abstract

Recently, the amount of multi-media data, such as movies, has been increasing rapidly. We often encounter situations which require fast and approximate retrieval of movies. A video is composed of multiple still images(frames). In this paper, we identify videos in real-time by performing retrieval of still images successively. When we use hierarchical similarity search indexes, results that are far away from a query are more costly to obtain than results that are close to a query. Far results are not useful for video identification. Nevertheless, it is important to know that no close results exist to correctly identify a video. The similarity between consecutive images is usually high because of the nature of video data. In this paper we propose a type of cache that exploits this fact by skipping queries that are likely not to have close results. We also employ an early termination strategy that avoids unnecessary distance computations to retrieve far results while preserving the quality of the video identification. By conducting experiments, we show that both methods provide considerable improvements. The proposed caching technique is able to skip up to 40% of the queries. The early termination strategy is able to reduce the number of distance computations to 0.5%. The combination of both techniques provides even greater gains.