NAR-DIFF: A Noise-Adaptive Reflectance Diffusion Model for Low-Light Image Enhancement

Low-light image enhancement (LLIE) has long been a challenging problem due to the complexities of noise and illumination variations in low-light environments. In this paper, we propose an innovative method that combines Retinex theory and diffusion models to enhance low-light images. Our method decomposes the input image into reflectance and illumination components, where the reflectance is denoised using an adaptive diffusion-based reflectance denoising module, and the illumination is enhanced via a newly designed illumination enhancement module. Notably, we introduce an adaptive noise estimation network to guide the denoising process of the diffusion model, effectively removing noise while preserving image details. Extensive experiments on multiple benchmark datasets demonstrate that our method significantly outperforms existing techniques in terms of PSNR and SSIM, while reducing artifacts and preserving natural colors.