Scholarlink

<p>The integration of multimodal image data, particularly through visible and infrared image fusion, plays a crucial role in enhancing scene understanding for various applications. Traditional fusion methods predominantly rely on spatial domain features, often neglecting the frequency domain, which results in the loss of high-frequency details and textures. To address these limitations, we propose DDCFusion, a novel network which combines dual-domain feature learning through Laplacian pyramid encoding and wavelet transform convolution. The bimodal feature module (BIFM) enables efficient dual-domain feature extraction, while the dual-domain collaborative attention mechanism (DDAM) dynamically balances feature saliency across domains, ensuring optimal fusion quality. Experiments demonstrate that DDCFusion achieves superior performance in fusion quality metrics, with improvements of 11.5% in MI and 7.9% in <InlineEquation ID="IEq1"> <EquationSource Format="TEX">\({Q}_{ab/f}\)</EquationSource> <EquationSource Format="MATHML"><math> <msub> <mi>Q</mi> <mrow> <mi>a</mi> <mi>b</mi> <mo stretchy="false">/</mo> <mi>f</mi> </mrow> </msub> </math></EquationSource> </InlineEquation> compared to other methods. Target detection experiments validate DDCFusion’s high quality visual effects and adaptability to downstream tasks. This code is available at <a href="https://github.com/WHaoXiang/DDCFusion.git">https://github.com/WHaoXiang/DDCFusion.git</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

DDCFusion: enhancing visible–infrared image fusion via dual-domain collaborative learning

  • 期刊论文

  • 发表日期：

    2026年1月25日

查看全文
  • HaoXiang Weng,
  • Bo Chen,
  • Linjing Li,
  • Hua Zhang,
  • Kaiming Cao,
  • Xueci Xu,
  • Hongrui Miao

摘要

The integration of multimodal image data, particularly through visible and infrared image fusion, plays a crucial role in enhancing scene understanding for various applications. Traditional fusion methods predominantly rely on spatial domain features, often neglecting the frequency domain, which results in the loss of high-frequency details and textures. To address these limitations, we propose DDCFusion, a novel network which combines dual-domain feature learning through Laplacian pyramid encoding and wavelet transform convolution. The bimodal feature module (BIFM) enables efficient dual-domain feature extraction, while the dual-domain collaborative attention mechanism (DDAM) dynamically balances feature saliency across domains, ensuring optimal fusion quality. Experiments demonstrate that DDCFusion achieves superior performance in fusion quality metrics, with improvements of 11.5% in MI and 7.9% in \({Q}_{ab/f}\) Q a b / f compared to other methods. Target detection experiments validate DDCFusion’s high quality visual effects and adaptability to downstream tasks. This code is available at https://github.com/WHaoXiang/DDCFusion.git.