論文

査読有り
2015年

Experimental evaluation of multi-key content-based image retrieval

Smart Innovation, Systems and Technologies
  • Hideki Sato
  • ,
  • Shigemi Nagata

40
開始ページ
313
終了ページ
324
記述言語
英語
掲載種別
研究論文(国際会議プロシーディングス)
DOI
10.1007/978-3-319-19830-9_28
出版者・発行元
Springer Science and Business Media Deutschland GmbH

A user of existing Content-Based Image Retrieval (CBIR) systems suffers from single query key selection, in case that he/she might begin retrieval process toward an ambiguous goal. In the other case, he/she cannot judge whether one image is more adequate than the other exactly for retrieval. To overcome the difficulty, this paper proposes Multi-Key CBIR (MK-CBIR) which differs from Single-Key CBIR (SK-CBIR) in retrieving similar images from image databases with a multikey query, not a single-key one. To implement MK-CBIR, skyline query based MKCBIR (Φsky) and aggregate k-Nearest Neighbor query basedMK-CBIR (Ωsum, Ωmax, andΩmin) are presented. Experimental results on retrieval performance ofMK-CBIR are as follows: ∙ Recall of Ωmax is over 0.3 higher than that of k-NN search and the highest among those of MK-CBIR. It is comparatively stable against noisy keys. ∙ Recall of Ωmin is roughly the same as that of k-NN search and the lowest among those of MK-CBIR. ∙ Recall of Ωsum and recall of Φsky are roughly the same and between those ofΩmin and Ωmax. It is concluded that MK-CBIR is better than SK-CBIR in retrieval performance.

リンク情報
DOI
https://doi.org/10.1007/978-3-319-19830-9_28
ID情報
  • DOI : 10.1007/978-3-319-19830-9_28
  • ISSN : 2190-3026
  • ISSN : 2190-3018
  • SCOPUS ID : 84947929910

エクスポート
BibTeX RIS