Abstract
Infrared and visible image fusion aims to enhance scene perception by exploiting complementary information from the two modalities. However, existing methods mainly preserve infrared targets and visible texture details but struggle to generate natural colors under low-light or grayscale visible conditions. Although post-fusion colorization improves visual appearance, it may weaken infrared thermal responses and reduce target saliency. To address these issues, this paper proposes RACFusion, an infrared and visible image fusion method that integrates residual attention with colorization enhancement. First, the colorization-enhanced visible image is used as a pre-fusion color and texture prior and jointly fused with the infrared image to complement visible appearance with infrared thermal information. Then, a spatial channel attention module, a gradient residual block, and an infrared saliency-guided gating mechanism are introduced to enhance salient regions, edges, and textures while adaptively regulating cross-modal feature contributions. Finally, maximum intensity and gradient response constraints are employed to preserve both salient infrared targets and visible structural details. Subjective and objective evaluations on the TNO, LLVIP, VIFB, and MSRS datasets demonstrate that RACFusion achieves a better balance between natural color representation, infrared target saliency, and texture detail preservation than representative methods.
| Original language | English |
|---|---|
| Article number | 115063 |
| Journal | Pattern Recognition |
| Volume | 183 |
| DOIs | |
| Publication status | Published - Mar 2027 |
| Externally published | Yes |
Keywords
- Attention mechanism
- Colorization
- Gradient residual
- Image fusion
Fingerprint
Dive into the research topics of 'RACFusion: A residual attention and colorization enhancement approach for infrared and visible image fusion'. Together they form a unique fingerprint.Cite this
- APA
- Author
- BIBTEX
- Harvard
- Standard
- RIS
- Vancouver