A Robust and Fast Non-Local Means Algorithm for Image Denoising

《A Robust and Fast Non-Local Means Algorithm for Image Denoising》由会员上传分享，免费在线阅读，更多相关内容在学术论文-天天文库。

1、万方数据LiuYL，wangJ，Chenxetn1．Arobustandfastnou-Iocalmeansalgorithmforimageden。ising·JOURNALOFCOM—PUTERSCIENCEANDTECHNOLOGY23(2)：270—279Mar．2008ARobustandFastNon—LocalMeansAlgorithmforImageDenoisingYa，n—LiLiul，2(刘艳丽)，JinWan91(王进)，XiChenl(陈曦)，Yan—WenGu03(郭延文)andQ

2、un-ShengPen91，2(彭群生)1StateKeyLabofCAD&CG，ZhejiangUniversity，Hangzhou310058，China2DepartmentofMathematics}ZhejiangUniversity，Hangzhou310058，China3StateKeyLabofNovelSoftwareTechnology，NanjingUniversity}Nanjing210000，ChinaE—mail：liuyanli@cad．zju．edu．cuRevisedJa

3、nuary7，2008AbstractInthepaper，weproposearobustandfastimagedenoisingmethod．TheapproachintegratesbothNon—LocalmeansalgorithmandLaplacianPyramid．Givenauimagetobedenoised，wefirstdecomposeitintoLaplacianpyramid．ExploitingtheredundancypropertyofLaplacianpyramid，we

4、thenperformnon。localmeansoneverylevelimageofLaplacianpyramid．Essentially．weusethesimilarityofimagefeaturesinLaplacianpyramidtoacta8weighttodenoiseimage．SincethefeaturesextractedinLaplacianpyramidarelocalizedinspatialpositionandscale，theyaremuchmoreabletodesc

5、ribeimage，andcomputingthesimilaritybetweenthemismorereasonableandmorerobust·Also,basedoiltheemcientSummedSquareImagefSSI)schemeandFastFourierTransform(FFT)，wepresentanacceleratingalgorithmtobreakthebottleneckofnon—localmeansalgorithm—similaritycomputationofc

6、omparewindows．Afterspeedup，ouralgorithmisfiftytimesfasterthanoriginalnon—localmeansalgorithm．Experimentsdemonstratedtheeffectivenessofouralgorithm．Keywordsimagedenoising，non—localmeans，Laplacianpyramid，summedsquareimage)FFT1IntroductionWiththeprevalenceofdig

7、itMcamerasandscan-ners，digitalimagescanbeeasilyacquirednowadays．Unfortunately，digitalimagesareoftendegradedbynoiseduringacquisitionortransmissionprocess．Var—iOUSimage—relatedapplications，suchasmedicalim—ageanalysis，imagesegmentation，andobjectdetec—tion，etc．，

8、generallyrequireeffectivenoisesuppressionmethodtofurtherproducereliableresults．Therefore，imagedenoisinghasbeenoneofthemostimportantandwidelystudiedproblemsinimageprocessingandcomputervision．Sofa