正则化恢复联合稀疏表示的图像超分辨率重构.pdf

《正则化恢复联合稀疏表示的图像超分辨率重构.pdf》由会员上传分享，免费在线阅读，更多相关内容在行业资料-天天文库。

1、第40卷第9期光电工程V_0l_40．No．92013年9月Opto—ElectronicEngineeringSept，2013文章编号：1003—501X(2013)09—0001—07正则化恢复联合稀疏表示的图像超分辨率重构路锦正，吴斌，张启衡2(1．西南科技大学信e,I程学院，四川绵阳621010；2．中国科学院光电技术研究所，成都610209)摘要：为改善强降质图像的分辨率水平，提出了一种正则化恢复联合稀疏表示的单帧图像超分辨率重构框架。为同时放大图像并抑制模糊及噪声，首先根据退化估计正则化平衡极小问题的逼

2、近项和先验项，然后基于初步的锐利清晰图像和预先建立的图像超完备稀疏表示字典实现边缘保持的图像分辨率放大。正则化恢复的输出改善了传统学习法图像超分辨中低频分量的双立方插值版本，同时对降质的有效抑制降低了字典原子对退化信息的依赖性实验结果表明，本方法可对模糊含噪的低分辨率图像实现有效的超分辨率重构。关键词：正则化恢复；稀疏表示；学习法超分辨；超分辨率重构中图分类号：TP391．4文献标志码：Adol：10．3969~．issn．1003．501X．2013．09．001ImageSuper-resolutionRecon

3、structionwithRegularizati0nRestorationandSparseRepresentatiOnLUJinzheng，WUBin，ZHANGQiheng(1．SchoolofInformationandEngineering，SouthwestUniversityofScienceandTechnology，Mianyang621010，SichuanProvince，China；2．InstituteofOpticsandElectronics，ChineseAcademyofScienc

4、es，Chengdu610209，China)Abstract：Inordertoimproveresolutionofsingleflameimagewithseveredegradation，weproposeanovelsuper-resolutionreconstructionframeworkviaregularizationrestorationcombinedwithlearning—basedsparserepresentationenhancement．Toachieveenlargementand

5、suppressionofblurringandnoisesimultaneously,wecarefullybalancethedatafidelityandtheprioritemusingregularizationparameteronthebasisofverisimilarestimationofdegradation．Basedontheacquiredrelativelycleanimageandpre—constructedover-completesparserepresentationdicti

6、onary,imageresolutionzoomingwithcharacteristicofedge—preservingCallthenberealized．Fundamentally,theoutputofprecedingregularizationreversionremarkablybetterslow-frequencyqualityofbicubicinterpolationversioninconventionallearning—basedsuper-resolutionFurthermore，

7、theriddingofblurandnoisecanfavorablyweakendependencyofatomstodegradedinformation．Consequently,theircombinationoftwotechniquescallremarkablyeliminateblurandnoise，andmeanwhile，removeannoyingedgeartifactsofenlargedimage．Experimentresultsdemonstratethattheaddressed

8、approachproducesvisuallypleasingmagnificationforblurryandnoisylow-resolutionimage．Keywords：regularizationrestoration；sparserepresentation；learning—basedsuper-resolution；supe