Multi-domain-based Automatic Image Representation Using Semantic Tolerance Relation Models

The nature of the concepts regarding images in many domains are imprecise, and the interpretation of finding similar images is also ambiguous and subjective on the level of human perception. To solve these problems, in this paper, images´ semantic categories and the tolerance degree between them are defined systematically, and the machine learning-based approach of modeling tolerance relations between the semantic classes is proposed. Furthermore, the method of the semantic tolerance relation model-based image representation and the corresponding image semantic categorization algorithm is also presented. We apply the proposed approach to the representations of images regarding the nature vs. man-made domain, human vs. non-human domain, and temporal domain, and compare the categorization results of them with the results not using semantic tolerance relation model. The results show the effectiveness of proposed method.