Rethinking Low-Light Image Enhancement: A Log-Domain Intensity--Chromaticity Decoupling Perspective

2026-05-04

AI Summary

This work addresses color distortion and detail degradation in low-light image enhancement caused by channel-wise abnormal amplification and chrominance noise. To mitigate these issues, the authors propose a novel paradigm that decouples intensity and chrominance components in the logarithmic domain. By explicitly decomposing the image into intensity and chrominance representations in log space and applying separate reconstruction constraints to each, the method effectively suppresses inter-channel imbalance and chrominance noise. Evaluated on the LOLv2-Real dataset, the approach achieves a PSNR of 29.71 dB and an SSIM of 0.89, significantly outperforming existing state-of-the-art methods. Furthermore, it demonstrates notable improvements in downstream low-light face detection performance, underscoring its practical efficacy.

Abstract

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.

Problem

• Low-Light Image Enhancement
• Chromatic Noise
• Channel Amplification
• Downstream Face Detection

Innovation

• Log-Domain Decoupling
• Intensity-Chromaticity Separation
• Low-Light Image Enhancement
• Chromatic Noise Suppression
• Reconstruction Constraints