Explicit reconstruction constraints derived from the decoupled representation are further imposed to suppress abnormal channel amplification and chromatic noise. Experiments on LOLv2-Real, MIT-Adobe FiveK, and LSRW show that the proposed method achieves competitive or superior quantitative and visual performance, reaching 29.71 dB PSNR and 0.89 SSIM on LOLv2-Real. DarkFace experiments further indicate improved downstream face detection under low-light conditions. Code and pretrained models are available at: https://github.com/mubaisam/ICD.
@article{arxiv.2605.02627,
title = {Rethinking Low-Light Image Enhancement: A Log-Domain Intensity--Chromaticity Decoupling Perspective},
author = {Guangrui Bai and Yifan Mei and Yahui Deng and Yuhan Chen and Yuze Qiu and Wenhai Liu and Erbao Dong},
journal= {arXiv preprint arXiv:2605.02627},
year = {2026}
}