Unsupervised Multimodal Hashing for Cross-modal retrieval.

With the advantage of low storage cost and high efficiency, hashing learning has received much attention in the domain of Big Data. In this paper, we proposed a novel unsupervised hashing learning method to cope with the situation where massive unlabeled data is obtained easily for the era of big data. Our method aims to learn the robust hashing functions by solving how to directly preserve the local manifold structure by hash codes. To address this problem, both the local semantic structure in textual space and the geometric structure in the visual space is explored to learn compact hash codes. Besides, the $\ell_{2,1}$-norm constraint is imposed on the projection matrices to select relevant and discriminative features from different modal spaces simultaneously. Extensive experiments are performed to evaluate the proposed method, termed Unsupervised Multimodal Hashing (UMH), on the three publicly available datasets and the experimental results show that the proposed UMH can achieve superior performance over the state-of-the-art methods.