Multi-label enhancement based self-supervised deep cross-modal hashing

Deep cross-modal hashing which integrates deep learning and hashing into cross-modal retrieval, achieves better performance than traditional cross-modal retrieval methods. Nevertheless, most previous deep cross-modal hashing methods only utilize single-class labels to compute the semantic affinity across modalities but overlook the existence of multiple category labels, which can capture the semantic affinity more accurately. Additionally, almost all existing cross-modal hashing methods straightforwardly employ all modalities to learn hash functions but neglect the fact that original instances in all modalities may contain noise. To avoid the above weaknesses, in this paper, a novel multi-label enhancement based self-supervised deep cross-modal hashing (MESDCH) approach is proposed. MESDCH first propose a multi-label semantic affinity preserving module, which uses ReLU transformation to unify the similarities of learned hash representations and the corresponding multi-label semantic affinity of original instances and defines a positive-constraint Kullback–Leibler loss function to preserve their similarity. Then this module is integrated into a self-supervised semantic generation module to further enhance the performance of deep cross-modal hashing. Extensive evaluation experiments on four well-known datasets demonstrate that the proposed MESDCH achieves state-of-the-art performance and outperforms several excellent baseline methods in the application of cross-modal hashing retrieval. Code is available at: https://github.com/SWU-CS-MediaLab/MESDCH.