CN115249211B 一种基于水下非均匀入射光模型的图像复原方法 （复旦大学）

一种基于水下非均匀入射光模型的图像复本发明发明涉及一种基于水下非均匀入射2步骤S4，将所述处理最低频子带及所述处理高频子带进ccccc其中，基于所述水下非均匀入射光模型，得到非均匀入射光下介质J'c步骤S2_1，采用简单线性迭代聚类算法对所述最低频3步骤S3_4，将抑制噪声后的所述各高频子带除以对应的所述各高频子带的透射系数，4(RedDarkChannelPrior，RDCP)方法；文献[7]则报道了一种基于图像模糊和光吸收始应用于提升水下图像的质量，文献[10]就报道了一种卷积神经网络(CNN[0006][1]LinS,ZhaoY.Reviewonkeytechnologiesoftargetexplorationinunderwateropticalimages[J].Laser&OptoelectronicsProgress,2020,57(6):060002.[0007]林森,赵颍.水下光学图像中目标探测关键技术研究综述[J].激光与光电子学进5enhancementandrestorationmethods[J].JournalofImageandGraphics,2017,22[0010][3]AncutiC,AncutiCO,HaberT,etal.Enhancingunderwaterimagesandvideosbyfusion[C]//2012IEEEConferenceonComputerVisionandPatternRecognition(CVPR),June16_21,2012,Providence,RI.New[0011][4]McGlameryBL.Acomputermodelforunderwatercamerasystems[J].ProceedingsoftheSPIE,198prior[J].IEEETransactionsonPatternAnalysisandMachineIntelligence,2011,33(12):2341_2353.[0013][6]GaldranA,PardoD,PicónA,etal.Automaticred_channelunderwaterimagerestoration[J].JournalofVisualCommunicationandImageRepresentation,2015,26:132_145.[0014][7]PengYT,CosmanPC.Underwaterimagerestorationbasedonimageblurrinessandlightabsorption[J].IEEETransactionsonImageProcessing,2017,26(4):1579_1594.[0015][8]DaiC,LinM,WuX,etal.Singlehazyimagerestorationusatmosphericscatteringmodel[J].Signalprocessing,2020,166:107257.[0016][9]WangGL,TianJD,LiPY.Imagecolorcorrectionbasedondoubletransmissionunderwaterimagingmodel[J].ActaOpticaSinica,2019,39(9):0901002.usingDehazenetandHWD[J].JournalofMarineScienceandTechnology,2018,26(4):531_540.[0019][11]DaiCG,LinMX,Wchannelandfusionforunderwaterimageenhancement[J].ActaOpticaSinica,2018,38(11):1110003.[0021][12]LandEH,McCannJJ.Lightnessandretinextheory[J].JournaloftheOpticalSocietyofAmerica,1971,61(1):1_11.[0022][13]LiuX,ZhangH,CheungYM,etal.Efficientsingleimagedehazinganddenosing:anefficientmulti_scalecorrelatedwaveletapproach[J].ComputerVisionandImageUnderstanding,2017,162:23_33.[0023][14]DuarteA,CodevillaF,GayaJDO,etal.Adatasettoevaluate6underwaterimagerestorationmethods[C]//OCEANS2016_Shanghai,April10_13,2016,Shanghai,China:IEEE,2016:1_6.[0024][15]DonohoDL.De_noisingbysoft_thresholding.IEEETransactionsonInformationTheory,1995,41(3):613_627.algorithmbasedontetrolettransform[J].ActaOpticaSinica,2017,37(9):0910002.[0027][17]AchantaR,ShajiA,SmithK,etal.SLICSuperpixelsComparedtoState_of_the_ArtSuperpixelMethods[J].IEEETransactionsonPatternAnalysisandMachineIntelligence,2012,34(11):2274_2282.[0028][18]YangM,LiuJ,LiZ.Superpixel_BasedSingleNighttimeImageHazeRemoval[J].IEEETransactionsonMultimedia,2018,20(11):3008_3018.[0029][19]ZhouW,LiD,ZhangJQ.Acomplementarycooncolorimagesharpnessassessmentforautofocus.ActaAutomaticaSinica,2020,46(8):1615_1627.[0031][20]YangHT,ChangYL,WangJ,etal.ANewAutomaticExposureAlgorithmforVideoCamerasUsingLuminanceHistogram[J].ActaOpticaSinica,2841_848.[0033][21]YangAP,WangJB,YangBW,etal.JointDeepDenoisingPriorforImageBlindDeblurring[J].ActaOpticaSinica,2018,38(10):1010003.[0034]杨爱萍,王金斌,杨炳旺,等.联合深度去噪先验图像盲去模糊[J].光学学报,2018,38(10):1010003.[0035][22]SheikhHR,WangZ,CormackL,etal.LIVEimagedatabaserelease2(2005)[Online],available:/research/quality/,September1,2011.[0036][23]GonzalezRC,RichardEW.Digitalimageprocessing[M].RuanQQ,RuanYZ,Transl.2thedition.Beijing:ElectronicIndustryPress,2007:72_73.[0038][24]AncutiCO,AncutiC,VleeschouwerCD,etal.Colorbalanceandfusionforunderwaterimageenhancement[J].IEEETransactionsonImageProcessing,2018,27(1):379_393.[0039][25]LiCY,GuoCL,RenWQ,etal.Anunderwaterimageenhancement7benchmarkdatasetandbeyond[J].IEEETransactionsonImageProcessing,2020,29:4376_4389.[0040][26]YangM,SowmyaA.Anunderwatercolorimagequalityevaluationmetric[J].IEEETransactionsonImageProcessing,2015,24(12):6062_6071.[0041][27]SaraU,AkterM,UddinMS.ImagequalityassessmentthroughFSIM,SSIM,MSEandPSNR—acomparativestudy[J].JournalofComputerandCommunications,2019,07(3):8_18.cc8声与散射均得到抑制的图像J'c(x)的局部梯度，tΩ表示局部小窗口内的透射系数，增强的高频子带。[0067]根据本发明所涉及的基于水下非均匀入射光模型的图像复原方法，包括如下步9[0074]图7是本发明的测试例一中参考图像及其对应的浑浊水下图像的傅里叶变换频谱合实施例及附图对本发明基于水下非均匀入射光模型的图像复原[0078]下述实施例中基于水下非均匀入射光模型的图像复原方法中的非均匀入射光模[0087]由于光在水下传播的过程中，水体对不同光谱的吸收衰减程度是随水体浑浊程[0093]如图1所示，本发明实施例中基于水下非均匀入射光模型的图像复原方法包括如cc低频子带做超像素分割(SuperpixelSegmentation)，即将相邻且特征相似的像素点分割[0118]步骤S2_2，利用环境光Bc_lp及最低频子带Ic_lp(x)得到最低频子带的透射系数tlp声与散射均得到抑制的图像J'c(x)的局部梯度，tΩ表示局部小窗口内的透射系数，及另外收集到累计约200张自然光照下高对比度无色偏图像进行了统计。通过随机观测发[0157]测试方法：假设水下介质散射引起的图像模糊也主要集[0159]图7是本发明的测试例一中参考图像及其对应的浑浊水下图像的傅里叶变换频谱[0180]CNN结合HWD算法虽然能得到高清晰度的图像，且进行边缘增强后其细节较为突[0182]选取水下彩色图像质量评价(UnderwaterColorImageQualityEvalua