When Optimal Transport Meets Photo-Realistic Image Dehazing With Unpaired Training

Image hazing is crucial for enhancing the image visibility and mitigating the weather degradations. However, most existing approaches rely on the paired hazy and clean images, which are challenging to obtain in real-world scenarios. To this end, we propose an oriented Bayesian-regularized consistent optimal transport (OBCOT) framework, which formulates the unpaired image dehazing task as an optimal transport (OT) problem. Specifically, we introduce a structure-preserving transport cost, incorporating the structural similarity (SSIM) constraint to minimize the duality gap between the primal and dual formulations, while preserving the structural details of reconstructed images. Furthermore, we derive the Bayesian frequency-domain regularization (BFR) to balance the spectral consistency with clean References and repulsion from hazy patterns. In addition, we employ a pretrained one-step stable diffusion model as the restoration network, which is fine-tuned using the low-rank adaptation (LoRA) adapters and zero convolutional layers, while integrating the domain-specific text prompts for both degraded and clean images to guide the generation process. Extensive experiments demonstrate that our method surpasses the existing well-performing unpaired learning approaches, achieving notable improvements in both the fidelity and photo-realism.