基于多级残差跳跃连接网络的图像超分辨率重建

打开文本图片集

关键词：图像超分辨率重建；SRGAN；坐标注意力机制；多级残差跳跃连接网络；PatchGANDOI：10.15938/j. jhust.2025.02.008中图分类号：TP391 文献标志码：A 文章编号：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

图像超分辨率重建（image super-resolution re-construction，SR）技术是指将低分辨率（lowresolution，LR）图像通过特定的算法恢复成具有更好的视觉效果和更清晰细节的高分辨率（highresolution，HR）图像[1]，目前已被广泛应用于人脸识别、医学成像[2]、军事遥感[3]和图像视频处理[4]等领域

传统的图像超分辨率重建算法以基于机器学习方法为主，常见的重建方法可以分为以下3类：基于插值[5]、基于重建[和基于学习的方法。（剩余14100字）