Scholarlink

<p>With the rapid growth of multimodal data types, unsupervised cross-modal hashing retrieval technology has become a vital solution for efficient storage and fast retrieval by learning universal binary hash codes. However, existing methods often suffer from inaccurate similarity measurements and imbalanced information between modalities, which limit the further improvement of retrieval performance. To address these challenges, this paper introduces a Similarity-Optimized and Semantic-Aligned Method for Unsupervised Cross-modal Hashing Retrieval (SOSAH), which contains a similarity graph optimization strategy and a hierarchical multimodal semantic-aware alignment module. (1) The similarity graph optimization strategy enhances the similarity between closely related samples while suppressing the similarity of unrelated samples. This refinement improves the expressiveness of the similarity matrix, effectively guiding the learning of highly discriminative hash codes. (2) To solve the problem of the imbalance of the modalities, we construct a hierarchical multimodal semantic-aware alignment module. This module employs a graph convolutional network to align intra-modal correlations and a hierarchical similarity aggregation network to align inter-modal relationships. Experimental results on two publicly available datasets demonstrate the effectiveness of the proposed approach and show significant improvements in retrieval performance.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

A similarity-optimized and semantic-aligned method for unsupervised cross-modal hashing retrieval

  • 期刊论文

  • 发表日期：

    2025年8月8日

查看全文
  • Xin Li,
  • Xiuyuan Li,
  • Mingyong Li,
  • Mingyuan Ge

摘要

With the rapid growth of multimodal data types, unsupervised cross-modal hashing retrieval technology has become a vital solution for efficient storage and fast retrieval by learning universal binary hash codes. However, existing methods often suffer from inaccurate similarity measurements and imbalanced information between modalities, which limit the further improvement of retrieval performance. To address these challenges, this paper introduces a Similarity-Optimized and Semantic-Aligned Method for Unsupervised Cross-modal Hashing Retrieval (SOSAH), which contains a similarity graph optimization strategy and a hierarchical multimodal semantic-aware alignment module. (1) The similarity graph optimization strategy enhances the similarity between closely related samples while suppressing the similarity of unrelated samples. This refinement improves the expressiveness of the similarity matrix, effectively guiding the learning of highly discriminative hash codes. (2) To solve the problem of the imbalance of the modalities, we construct a hierarchical multimodal semantic-aware alignment module. This module employs a graph convolutional network to align intra-modal correlations and a hierarchical similarity aggregation network to align inter-modal relationships. Experimental results on two publicly available datasets demonstrate the effectiveness of the proposed approach and show significant improvements in retrieval performance.