Skip to Main Content
Effective information retrieval (IR) using domain knowledge and semantics is one of the major challenges in IR. We propose a framework that can facilitate image retrieval based on a sharable domain ontology and thesaurus. In particular, case-based learning (CBL) using a natural language phrase parser is proposed to convert a natural language query into resource description framework (RDF) format, a semantic-web standard of metadata description that supports machine readable semantic representation. This same parser also is extended to perform semantic annotation on the descriptive metadata of images and convert metadata automatically into the same RDF representation. The retrieval of images then can be conducted by matching the semantic and structural descriptions of the user query with those of the annotated descriptive metadata of images. We tested in our problem domain by retrieving the historical and cultural images taken from Dr. Ching-chih Chen's "First Emperor of China" CD-ROM (1991) as part of our productive international digital library collaboration. We have constructed and implemented the domain ontology, a Mandarin Chinese thesaurus, as well as the similarity match and retrieval algorithms in order to test our proposed framework. Our experiments have shown the feasibility and usability of these approaches.