Diffusion-Based Low-Light Image Enhancement with Color and Luminance Priors

TL;DR

This paper introduces a diffusion-based low-light image enhancement method that uses a structured control embedding module to incorporate physical priors, achieving state-of-the-art results across multiple datasets.

Contribution

The paper presents a novel diffusion framework with a structured control embedding module that decomposes images into informative components for guided enhancement, improving generalization and performance.

Findings

Achieves state-of-the-art quantitative and perceptual results

Generalizes well across multiple benchmark datasets

Trained solely on the LOLv1 dataset without fine-tuning

Abstract

Low-light images often suffer from low contrast, noise, and color distortion, degrading visual quality and impairing downstream vision tasks. We propose a novel conditional diffusion framework for low-light image enhancement that incorporates a Structured Control Embedding Module (SCEM). SCEM decomposes a low-light image into four informative components including illumination, illumination-invariant features, shadow priors, and color-invariant cues. These components serve as control signals that condition a U-Net-based diffusion model trained with a simplified noise-prediction loss. Thus, the proposed SCEM equipped Diffusion method enforces structured enhancement guided by physical priors. In experiments, our model is trained only on the LOLv1 dataset and evaluated without fine-tuning on LOLv2-real, LSRW, DICM, MEF, and LIME. The method achieves state-of-the-art performance in quantitative and perceptual metrics, demonstrating strong generalization across benchmarks. https://casted.github.io/scem/.