2009 年 21 巻 2 号 p. 184-193
The nature of the concepts regarding images in many domains are imprecise, and the interpretation of finding similar images is also ambiguous and diverse on the level of human perception. Considering these features, in this paper, images' semantic classes and the tolerance degree between them are defined systematically, and the approach of modeling tolerance relations between the semantic classes is proposed. On the basis of it, a general mechanism of representing images' semantics by associative values with predefined classes regarding a corresponding dimension is depicted. Moreover, as demonstration, the methods of generating associative values with defined classes regarding the nature vs. man-made dimension and human vs. non-human dimension are described, and experimental results of images' retrieval show the effectiveness of our proposed mechanism of representing images' semantics in improving the precision-recall performance.
