Latent semantic-enhanced discrete hashing for cross-modal retrieval

Hashing methods have been proposed for the cross-modal retrieval tasks due to their flexibility and effectiveness. The main idea of cross-modal hashing is to embed heterogeneous multimedia data into common Hamming space. How to effectively exploit the modal semantic information and reduce optimization loss have been a challenging problem for existing cross-modal hashing methods. To address these issues, we propose a supervised cross-modal hashing method, called Latent Semantic-Enhanced discrete Hashing (LSEH). LSEH first leverages matrix factorization to obtain individual latent semantic representations of different modalities, and then applies correlation analysis and kernel discriminant analysis when projecting the latent semantic representations into the common Hamming space. Finally, the binary codes are directly generated with discrete optimization strategy. Experimental results on four benchmark datasets demonstrate that LSEH outperforms state-of-the-art cross-modal hashing methods in terms of retrieval accuracy, especially when dealing with image to text retrieval task, using shorter hash codes to associate images and texts.