Scholarlink

<p>Infrared and visible image fusion (IVIF) aims to integrate thermal signatures and structural patterns into a single image suitable for both human perception and machine processing. However, many existing methods rely on empirically-crafted fusion rules, which lack interpretability and frequently exhibit poor adaptation to cross-modal feature variations. To address these issues, we propose an IVIF method based on correlation-driven fusion rules and a parameter-free attention module. Our approach explicitly models feature interactions through statistical correlation analysis, deriving adaptive fusion weights directly from cross-modal feature maps to ensure robustness without training. We also introduce a parameter-free attention module to adaptively enhance texture details without introducing additional trainable parameters. Experimental results on public datasets demonstrate the superiority of our method in detail retention and target highlighting. Quantitative evaluations show improvements in metrics such as average gradient (AG) and Visual Information Fidelity (VIF) by up to 12% and 18%, respectively, compared to state-of-the-art methods on the LLVIP dataset. The source code is available at: <a href="https://github.com/CharlesShan-hub/CPFusion.">https://github.com/CharlesShan-hub/CPFusion.</a></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

Enhanced infrared and visible image fusion via correlation-driven rules and parameter-free attention mechanism

  • 期刊论文

  • 发表日期：

    2026年1月12日

查看全文
  • Hongtian Shan,
  • Yifan Chen,
  • Xitian Lu,
  • Jiangrong Lin,
  • Lei Deng,
  • Mingli Dong,
  • Lianqing Zhu

摘要

Infrared and visible image fusion (IVIF) aims to integrate thermal signatures and structural patterns into a single image suitable for both human perception and machine processing. However, many existing methods rely on empirically-crafted fusion rules, which lack interpretability and frequently exhibit poor adaptation to cross-modal feature variations. To address these issues, we propose an IVIF method based on correlation-driven fusion rules and a parameter-free attention module. Our approach explicitly models feature interactions through statistical correlation analysis, deriving adaptive fusion weights directly from cross-modal feature maps to ensure robustness without training. We also introduce a parameter-free attention module to adaptively enhance texture details without introducing additional trainable parameters. Experimental results on public datasets demonstrate the superiority of our method in detail retention and target highlighting. Quantitative evaluations show improvements in metrics such as average gradient (AG) and Visual Information Fidelity (VIF) by up to 12% and 18%, respectively, compared to state-of-the-art methods on the LLVIP dataset. The source code is available at: https://github.com/CharlesShan-hub/CPFusion.