Query-based video event definition using rough set theory

  • Authors:

  • Kimiaki Shirahama;Chieri Sugihara;Yuta Matsuoka;Kuniaki Uehara

  • Affiliations:

  • Graduate School of Economics, Kobe University, 2-1, Rokkodai, Nada, Kobe, Japan;Graduate School of Engineering, Kobe University, 1-1, Rokkodai, Nada, Kobe, Japan;Graduate School of Engineering, Kobe University, 1-1, Rokkodai, Nada, Kobe, Japan;Graduate School of Engineering, Kobe University, 1-1, Rokkodai, Nada, Kobe, Japan

  • Venue:
  • EiMM '09 Proceedings of the 1st ACM international workshop on Events in multimedia
  • Year:
  • 2009

Quantified Score

Hi-index 0.00

Visualization

Abstract

Since events queried by a user range much widely, indexing a video archive with pre-defined events is impractical. Hence, "query-based event definition" is an essential technique where an event is defined from example videos provided by the user. In this paper, we introduce a novel query-based event definition method to cover a large variation of low-level features in the same event. Specifically, due to arbitrary video production techniques, shots of the same event contain significantly different low-level features. Thus, we assume that these shots are distributed in different subsets in a feature space. To extract such subsets, we apply "rough set theory" to example shots relevant to an event (positive examples) and irrelevant example shots (negative examples). Thereby, we can extract different subsets where positive or negative examples can be correctly classified by "decision rules" consisting of specific low-level features. In this process, to avoid extracting over-specialized decision rules, we distinguish important low-level features from unimportant ones using "Multiple Correspondence Analysis (MCA)". Finally, the video archive is searched based on decision rules. Experimental results on TRECVID 2008 video archive show the possibility of our method to achieve the wide coverage of each event.