Lightweight Infrared and Visible Image Fusion based on Adaptive Weighting and Edge Gating

In infrared and visible sensory fusion, the discrepancy in physical mechanisms between thermal radiation imaging and optical photometric imaging often leads to cross-modal signal heterogeneity, resulting in feature imbalance, thermal halo artifacts and local texture degradation. Existing methods often compromise between the saliency of infrared targets and the high-fidelity reconstruction of visible details. To address the challenge of cross-modal signal heterogeneity, we propose a progressive image fusion architecture that decouples the fusion process into the intensity, gradient, and structural domains, thereby achieving multi-domain coupling and coordinated integration of cross-modal information. In the intensity domain, inspired by the distinct fusion requirements of representative region types, we design a Mexican-hat wavelet adaptive weighting strategy to achieve the adaptive allocation of heterogeneous modal energy. In the gradient domain, to suppress infrared thermal halo spillover, we design a visible-gradient-guided soft edge (VGSE) gating mechanism. By using the visible luminance gradient as a structural prior, the proposed gating map attenuates infrared responses near high-gradient structural boundaries while preserving thermal information in homogeneous regions. In the structural domain, while maintaining the foundational intensity distribution, we utilize self-guided filtering for high-frequency detail enhancement to compensate for visible details. Comprehensive experiments on the $\mathrm{M^3FD}$, $\mathrm{Roadscene}$, and $\mathrm{MSRS}$ datasets demonstrate that our method not only exhibits favorable performance in terms of objective metrics and visual perception, but also avoids reliance on deep network inference during the fusion process. Thus, we provide a lightweight, high-fidelity, and deployment-oriented fusion scheme for edge photoelectric display terminals.