AFFusion: Atmospheric scattering enhancement and frequency integrated spatial-channel attention for infrared and visible image fusion

Infrared and visible image fusion (IVIF) seeks to generate fused images that combine rich texture details with distinct thermal radiation features by integrating and leveraging complementary information from multiple sources. However, existing fusion methods frequently neglect the challenges posed by illumination degradation and inaccurate color contrast, which arise due to light energy loss and light scattering during atmospheric transmission. To address these limitations, this study introduces an innovative IVIF framework, termed AFFusion, which integrates an atmospheric scattering physical model with a frequency-domain feature component. By accurately predicting and estimating two key physical parameters-the transmission map and atmospheric light-within the scattering model, AFFusion harnesses atmospheric scattering principles to produce enhanced visible images, thereby mitigating the adverse effects of energy attenuation and scattering. Furthermore, to resolve artifacts and texture loss caused by traditional atmospheric scattering models, AFFusion incorporates Fourier transform in conjunction with spatial and channel attention mechanisms to selectively amplify amplitude and phase features in the frequency domain, thereby enhancing texture fidelity and detail representation within the fused images. Comprehensive experimental evaluations demonstrate that AFFusion surpasses state-of-the-art methods in both qualitative and quantitative performance metrics, while also providing robust support for high-level visual tasks. The implementation code is publicly accessible at https://github.com/cici0206/AFFusion.