Retrieve images by understanding semantic links and clustering image fragments

The main obstacle to realize real semantic-based image retrieval is that semantic description of versatile images is difficult. The basic ideas of this paper are that the semantics of an object can be refined through top–down orthogonal semantic classification and that the semantics of an object can be reflected by the semantics of relevant objects and the semantic relationships between them. To reflect the semantic relationships between images, we propose a set of primitive semantic links as the enhancement of the hyperlinks connecting Web pages. The semantic link network is the natural extension of the hyperlink network so it can inherit the existing theory and method on hyperlink network. Based on the single semantic image established upon the orthogonal semantic space and the semantic link space, the proposed image retrieval approach enables users to obtain the semantic clustering of relevant images rather than a list of isolated images as the output of the current search engine and to browse images along semantic paths with the support of semantic link reasoning. Semantic matching and reasoning for realizing intelligent semantic image retrieval is based on graph operation and matrix operation. 2003 Elsevier Inc. All rights reserved.