An image semantic retrieval system design and realization

CBIR technology has achieved a lot progress. Now people realize that it is impossible to reach the high-level semantic by the low-level physical features only, there exits a large semantic gap between them. Image semantic retrieval technology is now a research hotspot. It refers in three respects including semantic representation model, semantic information building and semantic retrieval techniques. In this paper, we proposed a novel framework for image semantic annotation and retrieval. It employs a semantic network as the semantic representation model, and uses the combination of semantic keywords, linguistic ontology and low-level features as the semantic retrieval technique. Through several times of users´ relevance feedback, semantic information building is completed constantly and automatically. To speed up the growth of semantic network and get a balance annotation, semantic seeds and semantic loners are employed. The framework will result in a rich semantic network and better retrieval performance in retrieval process.