Abstract
Image retrieval (IR) research has been ongoing for sometime. Two major paradigms in IR are developed respectively: Keyword-based metadata image retrieval and content-based image retrieval. The low retrieval precision and the difficulty of formulating an exact feature query are the major drawbacks of these approaches. To overcome these drawbacks, we propose a semantic-based image annotation and retrieval approach. In other words, we annotate images to be retrieved with semantic tags in a standard and uniform representation (RDF) that are defined and derived from thesaurus and domain concepts called domain ontology in form of OWL, so that the information retrieval can be conducted to some extent at the abstract “semantic” level. We also integrated various techniques, such as semantic web, case-based reasoning and complex matching algorithms to establish the system. Further, we realized that it is a difficult goal to achieve a complete annotation, therefore, we used the technique of intelligent agents to designed an annotator guide agent (AGA), who could guide an annotator to decide what to annotate for an image in a more effective and coherent manner with suggesting critical properties and domain commonsense. We also devised conflict detection patterns based on different data, ontology at different inference levels and proposed the corresponding automatic conflict resolution strategies. Finally, we conducted several experiments to compare the performance of the semantic-based retrieval, AGA, and automatic conflict resolution. The experiments showed that the proposed method improved the performance significantly.