類似画像検索における印象情報の寄与について(コミュニケーション支援及び一般)

概要

論文の詳細を見る
In this paper, we describe the relationship between subjective information and a performance in an image retrieval system. Although, in general, images are retrieved by using image feature, the retrieved images are evaluated by a user according to his/her subjective decision, i.e., impression. Thus, the human impression information is needed to be introduced into the image retrieval system in order to improve the performance of the system. The problem is how much the impression information should be incorporated or not. We apply several methods of multivariate analysis to the image retrieval system of family crests, such as Quantification IV and Canonical Correlation Analysis with kernel trick, which incorporate both impression and image information at various ratio. By comparing the image retrieval results in the experiment, we argue the contribution of subjective impression information.
2007-05-17