基于多級殘差跳躍連接網(wǎng)絡(luò)的圖像超分辨率重建

打開文本圖片集

關(guān)鍵詞：圖像超分辨率重建；SRGAN；坐標(biāo)注意力機(jī)制；多級殘差跳躍連接網(wǎng)絡(luò)；PatchGANDOI：10.15938/j. jhust.2025.02.008中圖分類號：TP391 文獻(xiàn)標(biāo)志碼：A 文章編號：1007-2683（2025）02-0073-09

Abstract：Imagesuper-resolutionreconstructiontechnologycanconvertlow-resolutionimagesintohigh-resolutionimageswith higherpixeldensityndclearedetailsandplaysanimportantoleiniliaryandmedicalfields.imingatteproblemofisfiient processingoftexturedetailsandcolorrestorationdegreeinexistingimagesuper-resolutionreconstructionalgorithms，amulti-level residualskipcoectionetwork（MRSCN）basedoncordinateatentionmechanismisproposedandapliedtothe SRGANmodel to realizefullutilizatiooflow-resolutionimagefeatures.Itisusedtorecoverthedetailsoftheimageandtooptimizetheperceivedloss using Charbonnier loss and TVloss.This algorithm is tested on Set5，Set14，Bsd100 and Urban100 data sets for 4x super-resolution reconstruction.Compared withothercommonlyusedsuper-resolutionalgorithms，thisalgorithmcanbeterretaintexturedetailsduring imagereconstruction，esultinginclearerimagedetails，etervisualfectsandetivereductionofteumberofparametersnth network.Intermsofobjectiveevaluationindicators，theaveragevalueofPSNRandSIMincreasedbyO.503dBand0.0076 respectively compared with the original SRGAN.

Keywords;image super resolution reconstruction；SRGAN；coordinate atention mechanism；multi-level residual skipcoectior network；PatchGAN

0引言

圖像超分辨率重建（image super-resolution re-construction，SR）技術(shù)是指將低分辨率（lowresolution，LR）圖像通過特定的算法恢復(fù)成具有更好的視覺效果和更清晰細(xì)節(jié)的高分辨率（highresolution，HR）圖像[1]，目前已被廣泛應(yīng)用于人臉識別、醫(yī)學(xué)成像[2]、軍事遙感[3]和圖像視頻處理[4]等領(lǐng)域

傳統(tǒng)的圖像超分辨率重建算法以基于機(jī)器學(xué)習(xí)方法為主，常見的重建方法可以分為以下3類：基于插值[5]、基于重建[和基于學(xué)習(xí)的方法。（剩余14100字）