Abstract: Aiming at the problem of limited cross-domain transformation quality caused by inter-domain feature differences in the existing diffusion models in image defogging task, this study breaks through the limitation of the traditional conditional generation framework in modeling the distribution differences between foggy and clear domains and proposes a new defogging method based on the Brownian Bridge diffusion model. A new transformation framework based on the Brownian Bridge diffusion model was proposed, and the image transformation between the foggy domain and the clear domain was modeled as a random Brownian bridge process. By designing a bidirectional diffusion mechanism, a bidirectional diffusion trajectory between the fogged image and the clear image was constructed to achieve end-to-end joint learning of the feature distribution between the two domains. The experimental results show that this method demonstrates significant advantages in the cross-domain image transformation task. Its bidirectional diffusion mechanism effectively overcomes the inter-domain difference problem of traditional conditional generation models, achieving a peak signal-to-noise ratio of 37.26 dB, a structural similarity of 0.9813, and a natural image quality assessment of 2.67 on the RESIDE dataset. The Brownian Bridge diffusion model achieves the directional decoupling and precise alignment of fog degradation features and clear image features by constructing a bidirectional Brownian motion trajectory between the fog domain and the clear domain, significantly improving the image restoration accuracy in complex atmospheric scattering scenarios.