Traditional foggy images are synthesized via the Single Atmospheric Scattering Model (ASM), which neglects the multiple scattering and light transport processes inherent in real-world environments. Consequently, an intrinsic feature distribution shift exists between synthetic and real foggy images, constraining the generalization of defogging models. To bridge this gap, this paper proposes a fog synthesis method based on a Multiple Scattering Model (MSM). By finely modeling the scattering characteristics of real foggy weather, the proposed method improves the physical realism and feature consistency of synthetic foggy images. Experiments on real-world datasets, including OverwaterHaze, RTTS, and SevereFog, show that models retrained with our proposed MSM consistently outperform baselines trained on the ASM across all evaluation metrics. On average, our method achieves 52.6% and 28.6% reductions in FADE and HazDesNet, respectively, along with 11.3% and 15.9% decreases in NIQE and PI. These results demonstrate the effectiveness and generalization capability of our approach.
Yan et al. (Mon,) studied this question.