Unsupervised low-light image enhancement with joint physical modeling and a priori
LIU Lingfeng
LIANG Yuxue
HU Chengshuang
ZHANG Zhen
LIU Jianlang
SHAO Kaixin
CHEN Yong
Abstract：The training and enhancement effect of supervised learning relies on paired datasets,but most of the data sets are synthetic datasets,which are poorly recovered and weakly generalized when migrated to real images.Based on the above problems,this paper proposes an unsupervised low-light image enhance-ment model with a joint physical model and multiple priors.First,supervised pre-training is performed on the synthetic dataset to meet the demand of the model to the low illumination;second,the model is fur-ther unsupervised,trained,and optimized on the real dataset and jointly with multiple physical a priori knowledge to better adapt it to the actual low illumination situation.The experiments show that the de-tails and illumination of the images are restored while the dependence on paired datasets is somewhat elimi-nated.
Keywords：image processinglow-light image enhancementunsupervised learningphysical modelingphysical priori
Publication Date：2026-06-25
Online Publishing Date：2026-08-28(First online date of this platform, not the publication date of the document)
Pages：9( 357-365 )