FIVFusion: Fog-free infrared and visible image fusion

As an essential visual enhancement technique, infrared and visible image fusion aims to generate high-quality images that integrate rich texture details and salient targets from both modalities, particularly under adverse environmental conditions. However, existing fusion methods often struggle in foggy weather due to severe image degradation, failing to fully exploit the complementarities and differences between infrared and visible modalities. This paper introduces a novel fog-free infrared and visible image fusion method, termed FIVFusion, which addresses the limitations of current approaches by synergistically integrating image defogging and fusion processes. FIVFusion employs a dual-modal feature aggregation module (DFAM) and a dual-branch perception decoupling module (DPDM) to execute fusion through an aggregation-to-decoupling strategy. In the aggregation phase, DFAM converts dual-modal images into a modality-independent shared representation to extract complementary information. In the decoupling phase, DPDM utilizes an infrared contrast-aware block (ICB) and a visible physical-aware block (VPB) guided by physical priors to enhance high-contrast infrared features and fog-affected visible textures, respectively. The enhanced visible features provide feedback to the fused image, incorporating infrared information, and further improving visible image defogging. To evaluate the effectiveness of FIVFusion, we constructed a foggy weather infrared and visible image dataset, FOGIV, based on the atmospheric scattering model. We evaluated FIVFusion against 16 state-of-the-art fusion methods using six metrics, demonstrating its superior performance in foggy weather conditions. Extensive experimental results confirm that FIVFusion achieves superior fusion quality with clearer texture details and more prominent salient targets compared to existing methods. Our code is publicly available at https://github.com/XiangheBi/FIVFusion .