Search papers, labs, and topics across Lattice.
This paper introduces Haze-Noise Diffusion (HNDiff), a novel framework for image dehazing that incorporates the physics of haze formation into the diffusion process. By utilizing a haze-aware noise scheduler, HNDiff intelligently adjusts noise levels based on haze density, which enhances the quality of the restored images. Experimental results demonstrate that HNDiff significantly outperforms existing dehazing methods, achieving state-of-the-art results on benchmark datasets.
Grounding image dehazing in the physics of haze formation allows HNDiff to achieve unprecedented restoration quality, outperforming traditional methods.
Existing diffusion-based methods have recently made significant progress in image dehazing. However, they typically neglect the physics of haze formation and reconstruct clean images from pure Gaussian noise, thereby limiting their restoration potential. To address this issue, we propose Haze-Noise Diffusion (HNDiff), a novel diffusion framework that embeds the atmospheric scattering model as an inductive bias. By grounding diffusion in physical principles, HNDiff ensures that the restoration aligns more closely with underlying mechanisms of haze formation. In its forward process, we introduce joint haze-noise diffusion with a haze-aware noise scheduler, which progressively adds both haze and noise to an image. Essentially, the scheduler adapts noise levels according to haze density, meaning that regions with heavier haze receive stronger noise injection to encourage content generation, while clearer regions receive lighter noise to better preserve details, which directly links the forward degradation process with the physics of haze. In the reverse process, we then derive a physically consistent dehazing-denoising process that simultaneously removes haze and noise to restore a clean image in a manner aligned with the forward degradation process. To further enhance practicality, we propose Latent HNDiff, which compiles clean latent priors that can be seamlessly integrated into existing dehazing networks to boost performance. Extensive experiments show that our work significantly improves leading dehazing backbones and achieves state-of-the-art results on benchmark datasets. The project page is available at https://jin-ting-he.github.io/HNDiff .
NVIDIA