Deep Color Consistent Network for Low-Light Image Enhancement

Zhao Zhang, Huan Zheng, Richang Hong, Mingliang Xu, Shuicheng Yan, Meng Wang; Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2022, pp. 1899-1908

Low-light image enhancement focus on refining the illumination and keep naturalness to obtain the normal-light image. Current low-light image enhancement methods can well improve the illumination. However, there is still color difference between the enhanced image and the ground-truth image. To alleviate this issue, we therefore propose deep color consistent network (DCC-Net) to preserve the color consistency for low-light enhancement. In this paper, we decouple a color image to two main components, a gray image and a color hist histogram. Further, we employ these two components to guide the enhancement, where the gray image is used to generate reasonable structures and textures and the corresponding color histogram is beneficial to keeping color consistency. To reduce the gap between images and color histograms, we also develop a pyramid color embedding (PCE) module, which can better embed the color information to the enhancement process according to the affinity between images and color histograms. Extensive experiments on the LOL, DICM, LIME, MEF, NPE and VV demonstrate that DCC-Net can well preserve color consistency, and performs favorably against state-of-the-art methods.