Query by Example by Extracting Inductive Query Definitions Using Rough Set Theory

• Shirahama Kimiaki

  Graduate School of Economics, Kobe University

• Matsuoka Yuta

  Graduate School of Engineering, Kobe University

• Uehara Kuniaki

  Graduate School of System Informatics, Kobe University

Bibliographic Information

Other Title

• ラフ集合理論を用いたクエリの帰納的定義に基づく例示映像検索

Abstract

We propose a query-by-example method that can retrieve a variety of shots relevant to a query, but these shots contain significantly different features due to varied shooting techniques and settings. Thus, we use rough set theory to extract multiple classification rules that characterize different subsets of example shots. We elaborate on how to extract useful rules from only a small number of example shots provided by the user. We incorporate bagging and the random subspace method into rough set theory. The former is useful to extract rules that cover a variety of shots, and the latter is useful to avoid extracting rules that overfit the example shots. Finally, although our method needs counter example shots, they are not provided by the user. Therefore, we use partially supervised learning to collect counter example shots from shots other than example shots. Experimental results on TRECVID 2009 video data validate the effectiveness of our method.

Journal

• The Journal of The Institute of Image Information and Television Engineers

  The Journal of The Institute of Image Information and Television Engineers 66 (5), J124-J135, 2012

  The Institute of Image Information and Television Engineers

Keywords

• 例示映像検索
• ラフ集合理論
• バギング
• ランダムサブスペース
• 部分教師付き学習

Details

• CRID: 1390282680104308992
• NII Article ID: 130002111210
• DOI: 10.3169/itej.66.j124
• ISSN: 18816908
• ISSN: 13426907
• Web Site: https://www.jstage.jst.go.jp/article/itej/66/5/66_5_J124/_pdf
• Text Lang: ja
• Data Source:
  • JaLC
  • Crossref
  • CiNii Articles
  • KAKEN
• Abstract License Flag: Disallowed