第31卷第4期2023年8月Vol.31No.4Aug.2023电脑与信息技术ComputerandInformationTechnology文章编号:1005-1228(2023)04-0028-03基于空间变化模糊核估计的图像盲超分辨率重建李轩,韩佳睿(沈阳航空航天大学电子信息工程学院,辽宁沈阳,110136)摘要:针对退化模式已知的图像超分辨率算法对于复杂退化的真实场景重建效果并不好,并且现有的盲图像超分往往假设模糊核空间不变,但在现实图像中的模糊核通常是空间可变的等问题。文章提出了一种基于空间变化模糊核估计的图像盲超分辨率网络(SpatiallyVariantKernelEstimationSuper-ResolutionNetwork,SKESRNet),SKESRNet由三部分构成:特征提取模块、模糊核重构模块和图像重建模块。给定任意模糊下的低分辨率图像,该网络首先利用特征提取模块得到输入图像的特征,然后根据得到的特征,利用模糊核重构模块估计出图像的模糊核,最后在图像重建模块完成输入图像的超分辨率重建。文章在多个基准数据集上进行了实验,结果表明该网络优于同类的图像盲超分辨率重建网络。关键词:模糊核估计;图像盲超分辨率;卷积神经网络中图分类号:TP391.41文献标识码:AImageBlindSuper-resolutionReconstructionBasedonSpatialVariationFuzzyKernelEstimationLIXuan,HANJia-rui(SchoolofElectronicInformationEngineering,ShenyangUniversityofAeronauticsandAstronautics,Shenyang110136,China)Abstract:Theimagesuper-resolutionalgorithmknownforthedegradationmodeisnotgoodforthereconstructionofcomplexdegradedrealscenes,andtheexistingblindimagesuper-resolutionoftenassumesthatthefuzzykernelisspatiallyinvariant.However,thefuzzykernelintherealimageisusuallyspatiallyvariable.Inthispaper,animageblindsuper-resolutionnetwork(SKESRNet)basedonspatialchangefuzzykernelestimationisproposed.SKESRNetincludesthreeparts:featureextractionmodule,fuzzykernelreconstructionmoduleandimagereconstructionmodule.Giventhelowresolutionimageunderarbitraryblur,thenetworkfirstusesthefuzzyfeatureextractionsubnetworktoobtaintheestimatedfeaturesoftheinputimage,thenusesthefuzzykernelreconstructionsubnetworktoestimatethefuzzykerneloftheimageaccordingtotheobtainedfeatures,andfinallyusestheimagereconstructionsubnetworktocompletethesuper-resolutionreconstructionoftheinp...