基于深度學(xué)習(xí)和Retinex理論的圖像增強(qiáng)方法

打開文本圖片集

引用格式：.基于深度學(xué)習(xí)和Retinex理論的圖像增強(qiáng)方法[J].現(xiàn)代電子技術(shù)，2025，48（（13）：36-42.

Image enhancement method based on deep learning and Retinex theory

WANG Dexu，WANG Zhifeng，YANWenqiang （Schoolof IntelligentManufacturingandControlEnginering，ShanghaiPolytechnicUniversity，ShanghaiO2oo，Chna）

Abstract：Thisstudyaimstoimproveimagequalityunderlow-lightconditions，primarilyaddressingissuesofreduced visibilityandcolordistortion.BycombiningimageprocesingtechniquesbasedonRetinextheoryandadvancedneuralnetwork algorithms，aninnovativeimageenhancementframeworkisproposed.Thisframeworkconsistsoftwoparts，namedanimage decompositionnetworkandanimageenhancementnetwork.The former isresponsiblefordecomposing theoriginal image into illuminationcomponentandreflectancecomponent，whilethelaterisresponsibleforoptimizing parametersandperformingY correctinbythenaturalimagequalityevaluator（NIQE），adjustingthebrightnessandcontrastoftheilluminationcomponent， andthenre-fusing theilluminationcomponent with thereflectancecomponent，soastoenhanceoverallimagequality.Testson standard datasetsLOLandLOL-V2showthatthe proposed methodoutperformsmost existing image enhancement algorithms in termsof thepeak signal-to-noiseratio （PSNR）andthestructural similarityindex measure （SIM），whichdemonstratesits effectiveness and practicalityin the field of low-light image enhancement.

Keywords：neural network;Retinex theory; NIQE; γ correction; PSNR; SSIM; image enhancement; image decomposition

0 引言

當(dāng)光線不足時，照片質(zhì)量往往會降低，表現(xiàn)為可見度減弱、色彩失真，以及細(xì)節(jié)模糊和不自然的色彩再現(xiàn)[1。（剩余12061字）