A Multimedia Information Retrieval Method Based on Cross-Modal Hashing

How to retrieve multimedia information effectively and efficiently is of great importance in information retrieval and computer vision. Therefore, in this paper, we aim to fully utilize the cross-modal hashing technology to solve the multimedia information retrieval problem. The main idea of this paper is to learn two different categories of hash functions for two different data modality, such as image modality and text modality. Then, the goal of this work is to learn several hashing functions, and then represent the samples to binary codes, and the proposed method generates hashing codes via connecting various modalities. In particular, we exploit the non-negative matrix factorization to learn a shared semantic space for different data modalities. Experimental results demonstrate that the proposed method can achieve higher accuracy of cross-domain multimedia information retrieval than other methods.