Advanced RetinexNet: A fully convolutional network for low-light image enhancement

Jiang Hai,Yutong Hao,Fengzhu Zou,Fang Lin,Songchen Han

DOI: https://doi.org/10.1016/j.image.2022.116916

2022-12-28

Abstract:Capturing images in weak illumination environments seriously degrades image quality, such as low visibility, low contrast, artifacts, and noise. Solving a series of degradation of low-light images can effectively improve the visual quality of the image and enhance the performance of high-level visual tasks. Based on the Retinex theory, we propose a novel fully convolutional network to simultaneously solve these degradations and effectively improve the visual quality of the image. The entire network contains two sub-networks, the Decom-Net and the Enhance-Net, the image is decomposed into illumination and reflectance through Decom-Net, while noise in the reflectance is suppressed through our proposed frequency-domain noise suppression loss, and the Enhance-Net is proposed to improve the contrast of the illumination map. Our method can properly improve the contrast and suppress noise at the same time. Extensive experiments demonstrate that our method outperforms the existing state-of-the-art methods by a large margin both quantitatively and visually. In addition, our results showed that the performance of the high-level visual task (i.e., face detection) can be effectively improved by using the enhanced results obtained by our method in low-light conditions.