论文 | Shuangli Du, Minghua Zhao, Yiguang Liu, Zhenzhen You, Zhenghao Shi, Jie Li:Low-light image enhancement and denoising via dual-constrained Retinex model

本文（Low-light image enhancement and denoising via  dual-constrained Retinex model）原载 Applied Mathematical Modelling，由四川大学刘怡光教授等科研人员创作，系四川大学智慧法治超前部署学科系列学术成果。后续会持续分享四川大学智慧法治超前部署学科系列学术成果，欢迎大家阅读。

Abstract:For low-light image enhancement, Retinex-based models have  been acknowledged as a representative technique, yet they frequently amplify noise  hidden in dark image. There have been some attempts to noise reduction. Because  they assume a uniform level of noise, the background seems blurry. In this paper, a  dual constraint is developed to perform Retinex decomposition and de-noising. First,  a patch-aware low-rank model (PALR) is proposed to reject noise. In contrast to the  global noise intensity assumption, PALR can measure noise intensity for each image  patch, and flexibly control the regularization extent to achieve a trade-off between  noise removal and details preservation. Second, noise looks like image background  texture instead of image structure. Relative total variation (RTV) is used to simulate  the visual difference and works well in enlarging image gradients for highlighting  details. By imposing PALR constraint on clean scene image, enforcing visual  difference constraint on the illumination and reflectance, a dual-constrained Retinex  9decomposition model (DCRD) is proposed, which can remove noise during  decomposition. DCRD can be solved by an alternating optimization algorithm.  Experiments on commonly tested low-light image datasets demonstrate the  competing performance of our proposed model in comparison with the  state-of-the-art methods.

Shuangli Du, Minghua Zhao, Yiguang Liu, Zhenzhen You, Zhenghao Shi,  Jie Li, Zhenyu Xu. Low-light image enhancement and denoising via  dual-constrained Retinex model，Applied Mathematical Modelling,Vol.116, pp.1-15, April 2023..（论文下载）