基于Retinex理论的低光图像增强算法  

Low-light image enhancement algorithm based on Retinex theory

作　　者：刘江山 苍岩[1] LIU Jiangshan;CANG Yan(College of Information and Communication Engineering,Harbin Engineering University,Harbin 150001,China)

机构地区：[1]哈尔滨工程大学信息与通信工程学院,黑龙江哈尔滨150001

出　　处：《应用科技》2024年第5期249-255,共7页Applied Science and Technology

基　　金：国家自然科学基金项目(61871142);中央高校基本科研业务费项目(3072020CFT0803)。

摘　　要：为解决低光图像存在的质量不佳和退化的问题,提出一种新的基于Retinex理论的低光图像增强网络用于低光图像的增强。与传统的基于Retinex的低光图像增强网络的三阶段处理方法不同,提出的网络一共包括分解网络和融合网络2个子网络,分别用于对低光图像的Retinex分解和对分解的反射分量与光照分量的融合,最后得到增强后的低光图像。改进后的2个子网络解决了低光图像存在的图像严重退化的问题,并通过与其他算法的对比试验验证了改进后的网络对低光图像增强效果的优越性。本文提出的低光图像增强算法能够明显降低低光图像增强过程中的噪声和颜色失真。A new low-light image enhancement network based on the theory of Retinex is proposed to address the issues of poor quality and degradation in low light images.Unlike the traditional three-stage processing method of Retinexbased low-light image enhancement networks,the network proposed in this paper consists of two sub-networks:a decomposition network and a fusion network.These two sub-networks are respectively used for Retinex decomposition of low-light images and fusion of the decomposed reflection and illumination components,resulting in enhanced lowlight images.The improved two sub-networks have solved the defects of severe image degradation in low light images,and the superiority of the improved network in enhancing low-light images has been verified through comparative experiments with other algorithms.The low-light image enhancement algorithm proposed in this paper can significantly reduce noise and color distortion during the low-light image enhancement process.

关 键 词：低光图像增强 图像退化 RETINEX理论 图像恢复 深度学习 图像处理 图像分解 图像融合 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]