Unsupervised Multi-modal Hashing for Cross-Modal Retrieval

The explosive growth of multimedia data on the Internet has magnified challenge information retrieval. Multimedia usually emerges in different modalities, such as image, text, video, and audio. Unsupervised cross-modal hashing techniques that support searching among multi-modal have gained importance large-scale retrieval tasks because advantage low storage cost high efficiency. Current methods learn hash functions by transforming high-dimensional into discrete codes. However, original manifold structure semantic correlation are not preserved well compact We propose a novel unsupervised method to cope with this problem from two perspectives. On one hand, textual space locally geometric visual reconstructed unified features seamlessly simultaneously. other \(\ell _{2,1}\)-norm penalties imposed projection matrices separately relevant discriminative experimental results indicate our proposed achieves an improvement 1%, 6%, 9%, 2% over best comparison four publicly available datasets (WiKi, PASCAL-VOC, UCI Handwritten Digit, NUS-WIDE), respectively. In conclusion, framework which combines learning multimodal graph embedding is effective codes superior performance compared state-of-the-art methods.