Scholarlink

With the proliferation of multimodal data, unsupervised cross-modal hashing has emerged as an efficient solution for large-scale retrieval. However, existing methods face challenges in accurate similarity measurement and modality imbalance. To address these limitations, we propose Semantic Enhanced Interaction Hashing (SEIH) with two key innovations: (1) a similarity matrix optimization strategy that improves measurement accuracy and (2) a multimodal semantic alignment module using graph convolutional networks and similarity aggregation to enhance both intra-modality consistency and inter-modality associations. Extensive experimental results show that our method has better retrieval performance on MIRFlickr-25K and NUS-WIDE multimodal benchmark datasets than other classical unsupervised cross-modal hashing methods.

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

Semantic Enhanced Interaction for Unsupervised Cross-Modal Hashing Retrieval

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Xin Li,
  • Mingyong Li

摘要

With the proliferation of multimodal data, unsupervised cross-modal hashing has emerged as an efficient solution for large-scale retrieval. However, existing methods face challenges in accurate similarity measurement and modality imbalance. To address these limitations, we propose Semantic Enhanced Interaction Hashing (SEIH) with two key innovations: (1) a similarity matrix optimization strategy that improves measurement accuracy and (2) a multimodal semantic alignment module using graph convolutional networks and similarity aggregation to enhance both intra-modality consistency and inter-modality associations. Extensive experimental results show that our method has better retrieval performance on MIRFlickr-25K and NUS-WIDE multimodal benchmark datasets than other classical unsupervised cross-modal hashing methods.