In order to improve the performance of the cross-media data retrieval, we propose a construction method of hashing function for cross-media retrieval on the basis of the traditional hash index. The method considers both Intra-modality and Inter-modality similarities, and obtains the similarity matrices by calculating the contact ratio of structured words. By establishing the Bayesian model, this method expresses a dependent relationship between variables which include the multimodal similarity matrix and the hamming coding matrix. Then it trains the hash function by utilizing parameter learning strategy, and maps data from high-dimensional space to hamming space, finally realizes the hamming code. The theory analysis and simulation results prove the efficiency and accuracy of the method.
展开▼
