Image Light Enhancement Model Based on Illuminance- Diffusion

Low-light image enhancement is a critical task that is essential for improving image quality and performance. In this paper, we rethink the low-light image enhancement task and propose a physically interpretable decomposition diffusion model for low-light image enhancement called Illuminance-Diffusion. Based on retinex theory as well as convolutional neural networks, we have designed an innovative network architecture, ReC. The network is able to efficiently decompose the input image into two key components: illuminance map and reflectance map. The illuminance map is then augmented with brightness using wavelet diffusion model to restore image brightness and reduce noise. Extensive experiments were conducted on the sports fields dataset, and the experimental results show that our model performs excellently in PSNR and SSIM.