Graph Rebasing and Joint Similarity Reconstruction for Cross-Modal Hash Retrieval

Abstract

Cross-modal hash retrieval methods improve retrieval speed and reduce storage space at the same time. The accuracy of intra-modal and inter-modal similarity is insufficient, and the large gap between modalities leads to semantic bias. In this paper, we propose a Graph Rebasing and Joint Similarity Reconstruction (GRJSR) method for cross-modal hash retrieval. Particularly, the graph rebasing module is used to filter out graph nodes with weak similarity and associate graph nodes with strong similarity, resulting in fine-grained intra-modal similarity relation graphs. The joint similarity reconstruction module further strengthens cross-modal correlation and implements fine-grained similarity alignment between modalities. In addition, we combine the similarity representation of real-valued and hash features to design the intra-modal and inter-modal training strategies. GRJSR conducted extensive experiments on two cross-modal retrieval datasets, and the experimental results effectively validated the superiority of the proposed method and significantly improved the retrieval performance.