A semantic-consistency asymmetric matrix factorization hashing method for cross-modal retrieval

Hashing methods have recently received widespread attention due to their flexibility and effectiveness for cross-modal retrieval tasks. However, existing cross-modal hashing methods have a common challenging problem, how to effectively exploit semantic information to learn discriminative hash codes while saving storage and computation cost. To address this issue, in this paper, we propose an efficient Semantic-consistency Asymmetric Matrix Factorization Hashing (SAMFH) method. Specifically, this method first leverages matrix factorization to obtain the latent semantic representations for different modalities and the label representation for class label information. To further utilize semantic information and learn discriminative binary codes, we adopt an asymmetric supervised learning strategy to fuse the pairwise semantic matrix into the framework. Finally, we directly update unified hash codes with an efficient discrete optimization strategy. Experimental results on three benchmark datasets demonstrate that our SAMFH method outperforms many state-of-the-art cross-modal hashing methods.

Automated summary

In this paper, an efficient Semantic-consistency Asymmetric Matrix Factorization Hashing (SAMFH) method is proposed to address the challenge of effectively exploiting semantic information for learning discriminative hash codes in cross-modal retrieval tasks.