基于Retinex理论和生成对抗网络的低照度图像增强算法

打开文本图片集

中图分类号：TP391.4 文献标识码：A 文章编号：2096-4706（2025）17-0058-04

Abstract：Aimingattheproblemsof insufficient brightness，noiseinterferenceanddetaillossof images inlow-light environment，this paper proposesalow-lightimageenhancementalgorithmbasedonRetinex theoryand GenerativeAdversarial Network （GAN）.The generatorof the algorithmconsistsofadecomposition network，an ilumination adjustment network and noisereductionblock.Firstly，thelow-lightimageisdecoupledintotheillminationcomponentandthereflectioncomponent bythedcompositionnetwork.Secondly，thebrightnessisorrectedbytheillminationadjustmentnetworkandthedetails areoptimizedbythenoisereductionblock.Finallyimagereconstructionisealizedbycomponentrecombination.Inadition， the algorithmalso introduces SEatention blockanddenseconnection block toenhance the feature propagationability.The experimentalresultsshowthatthealgorithmhasadvantages ineliminatingnoiseinterference，maintaining texturedetailsand improving multi-scene illumination adaptability.

Keywords： low-light image enhancement; Retinex theory; GAN; SE attention block; dense connection block

0 引言

在计算机视觉、数字摄影和安防监控等领域，图像质量直接影响后续任务的性能。（剩余6729字）