一种快速的超分辨率图像重构算法.pdf

《一种快速的超分辨率图像重构算法.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、电子·激光第24卷第2期2013年2月JournalofOptoelectr0nics·LaserVo1．24No．2February2013一种快速的超分辨率图像重构算法刘哲，张永亮，郝珉慧，张鹤妮(iNfirm2业大学理学院，陕西西安710129)摘要：基于稀疏表示的超分辨率(SR)图像重构算法需要求解ll范数优化问题，时间效率不高。本文在该方法的基础上加入平移不变性约束条件，并且考虑到稀疏参数对重构结果影响不大，对重构算法进行简化，避免了求解复杂的ll范数优化问题，实现了一种快速的图像SR重构。数值实验结果表明，本文算

2、法具有更高的时间效率。关键词：局部线性嵌入(LLE)；11范数优化；稀疏度；邻近数目；平移不变性中图分类号：TN911．73文献标识码：A文章编号：1005—0086(2013)02—0372'-06AfastsuperresolutionimagereconstructionalgorithmLIUZhe，ZHANGYong-liang，HAOMin-hui，ZHANGHe-ni(SchoolofScience，NorthwesternPolytechnicalUniversity，Xian710129，China)Abs

3、tract：Thesuperresolution(SR)reconstructionofimagecanbeachievedbythelocallylinearembed—dingalgorithmorthesparserepresentationalgorithm．ThelocallylinearembeddingalgorithmrequiresthenumberKofthenearestneighborsasapredefinedparameter．ImproperchoiceofKwillcausetheO—ver-

4、fittingorunder-fittingofdata，whichresultsinblurringofthereconstructedimage．Thesparserep—resentationalgorithmcouldselecttheneighborsadaptively，butneedstosolveacomplicatedandtimeconsuming11一normminimizingproblem．Inthiswork，anewsuperresolutionreconstructionalgorithmis

5、proprosedbyintroducingtheshiftinvarianceconstraints．SincethesparsityparameteraffectStherecon—structioninsignificantly，theproposedalgorithmissimplifiedtoavoidsolvingtheminimizingll—normproblem，whichleadstoasignificantreductioninthetimecomplexity．Thetimecomplexityoft

6、hepro—posedalgorithmis0(mr／)，whilethatofsolvingtheml’rn‘ml‘zm—gll—normbyLassois0(7r／3+2)．Ex—perimentalresultsindicatethattheproposedalgorithmismoreefficientthanpreviousalgorithms，andholdsagoodstabilityintimeconsumptionwiththeincreasingoverlapbetweencontiguousimagep

7、atchesandthedictionarysize．Thus，theproposedalgorithmcanbeimplementedwithabiggerdictionaryunderacceptabletime，whichcanbeadoptedfortherealtimesuperresolutionreconstructionapplications．Keywords：locallylinearembedding(LLE)；Z1一normminimizing；sparsity；neighbors；shiftinva

8、riance而现实的很多情况下，同一场景的多幅图像难以1引言获取，这就需要有效的单幅图像SR重构算法来解超分辨率(SR，superresolution)是利用给定的决～引。一幅低分辨率(LR，low—resolution)图像，重构出一传统的重构算法随着放大倍数的增大，重构幅同一场景的高分