Diff-Lighten: low-light image enhancement with retinex-based diffusion models

Image enhancement based on the Retinex algorithm under low illumination is an important research direction in the field of computer vision. Images captured under low light conditions often suffer from uneven illumination, severe noise, and significant loss of detail. Enhancing the quality of low-light images and addressing degradation issues not only help in advanced visual tasks such as object detection more effectively but also improves their performance in practical applications such as autonomous driving. This paper proposes a low-light image enhancement method based on Retinex image decomposition and diffusion models. The method leverages the excellent generative capabilities of diffusion models to achieve satisfactory results in detail and color restoration. We evaluated this method on the LOL dataset and demonstrated its effectiveness.