文档下载 文档下载导航
设为首页 | 加入收藏
搜索 请输入内容:  
 导航当前位置: 文档下载 > 所有分类 > IT/计算机 > 互联网 > Deep Belief Networks for Image Denoising
侵权投诉

Deep Belief Networks for Image Denoising

DeepBeliefNetworksforImageDenoisingMohammadAliKeyvanradMohammadPezeshki

Dept.ofComputerEngineeringandIT,Dept.ofComputerEngineeringandIT,AmirkabirUniversityofTechnology,AmirkabirUniversityofTechnology,

Tehran,IranTehran,Iran

keyvanrad@aut.ac.irm.pezeshki@aut.ac.ir

MohammadMehdiHomayounpour

Dept.ofComputerEngineeringandIT,

AmirkabirUniversityofTechnology,

Tehran,Iran

homayoun@aut.ac.ir

Abstract

DeepBeliefNetworkswhicharehierarchicalgenerativemodelsareeffectivetoolsforfeaturerepresentationandextraction.Furthermore,DBNscanbeusedinnu-merousaspectsofMachineLearningsuchasimagedenoising.Inthispaper,weproposeanovelmethodforimagedenoisingwhichreliesontheDBNs’abilityinfeaturerepresentation.Thisworkisbaseduponlearningofthenoisebehavior.Generally,featureswhichareextractedusingDBNsarepresentedasthevaluesofthelastlayernodes.WetrainaDBNawaythatthenetworktotallydistin-guishesbetweennodespresentingnoiseandnodespresentingimagecontentinthelastlaterofDBN,i.e.thenodesinthelastlayeroftrainedDBNaredividedintotwodistinctgroupsofnodes.Afterdetectingthenodeswhicharepresentingthenoise,weareabletomakethenoisenodesinactiveandreconstructanoise-lessimage.Insection4weexploretheresultsofapplyingthismethodontheMNISTdatasetofhandwrittendigitswhichiscorruptedwithadditivewhiteGaus-siannoise(AWGN).Areductionof65.9%inaveragemeansquareerror(MSE)wasachievedwhentheproposedmethodwasusedforthereconstructionofthenoisyimages.

1Introduction

Imagesignalsareoftencorruptedduetonoise.Removingnoisefromimageisanimportantissueincomputervision,becausethisstepcouldbethepreprocessingstepofmanyotherapplications.Uptonowvariousmethodshavebeenproposedtoremovenoise(denoise)fromvisualdata.FocusofmanyofthesemethodsisonFourierAnalysis[1],SpatialFiltering[2],andWaveletTransform

[3].AlsotherearesomeothermethodsbasedonSpareCodingandDictionaryLearning[4].Ontheotherhand,MachineLearningtoolssuchasConvolutionalNeuralNetworks(CNN)orDeepNeuralNetworks(DNN)havebeenusedinseveralpaperstotacklethisissue[5][6].Oneofthebiggestdif cultiesintrainingsuchdeepnetworksisthatthecostfunctionofsuchdeeparchitecturesgetsstuckinpoorlocaloptimaduetorandominitializationofweights.Hintonetal.[7]proposedanewgreedylayer-wisealgorithmtotacklethisissueandintroducedDeepBeliefNetworks(DBNs).DBNsareabletopresentagood”feature”representationofdata.Thesefeatureswhicharede nedasthepropertiesofinputdataarepresentedasnodesofthelastlayerofDBN.Asaresult,inthispaperwetrainaDBNsuchthatitlearnstoextractimagefeatures.ThetrainedDBNdistinguishesbetween”noisefeatures”and”cleanimagefeatures”inthelastlayerandpresentsthemintotwodistinctgroupsofnodes.Furthermore,DBNsarecapableofreconstructingtheinputdatabased

arXiv:1312.6158v2 [cs.LG] 2 Jan 2014

第1页

你可能喜欢

热门文档

相关文档

站点地图 | 文档上传 | 侵权投诉 | 手机版
本站所有资源均来自互联网,本站只负责收集和整理,均不承担任何法律责任,如有侵权等其它行为请联系我们.
文档下载 Copyright 2013 doc.guandang.net All Rights Reserved.  -  email