Associative image retrieval using knowledge in encyclopedia text

Applying existing keyword retrieval to image retrieval causes some problems. Among them, database developers must describe photograph contents in detail and adequate retrieval is difficult. These problems have been resolved by developing an associative retrieval technology using not keywords but semantic vectors as a retrieval method with an associative function such as a human being possesses. A semantic vector dictionary of more than 100,000 words was made from encyclopedia text. This paper explains the experimental image retrieval system for 36,000 photographs with the semantic vector dictionary made from the encyclopedia text. This system associates the words input by a user with the knowledge in the encyclopedia text and outputs ranked retrieval results. The effectiveness of this associative retrieval method is confirmed by evaluating content retrieval of images by a small benchmark and making an adaptive learning function of semantic vectors in the case in which a retrieval result is not the same as a user's subjective perspective would impose.