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.