欢迎来到天天文库
浏览记录
ID:33307869
大小:2.34 MB
页数:76页
时间:2019-02-24
《基于多特征信息融合的图像检索技术研究精选》由会员上传分享,免费在线阅读,更多相关内容在应用文档-天天文库。
1、华侨大学硕士学位论文基于多特征信息融合的图像检索技术研究姓名:叶宇光申请学位级别:硕士专业:计算机应用技术导教师:陈锻生20060915ABSTRACTContentBasedImageRetrieval(CBIR)hasbeenaveryactiveresearchareasince1990's.Inthecontentbasedimageretrieval,itisretrievedbysinglevisualcharacteratthebeginningoftheresearch,suchas:color,textureandfigurecharacters.C
2、olorandtexturecharacteraretwoveryimportantandwidelyusedimagecharactersinthecontentbasedimageretrieval.Sofirstlythesetwofeaturesarerespectivelydiscussedinthispaper.Andatthesametimeinvariabletheoryhasachievedsuccessfulapplicationinthecontentbasedimagedataretrievalinrecentyears,soinvariant
3、characterofimageretrievalisalsodiscussedinthispaper.First,twocolorspaces(RGBandHSV),threecolorhsitogramsmethods(traditionhsitogram,accumulationhsitogram,localaccumulationhsitogram),andcolormatchingalgorithmarediscussedinthispaper.Thentexturefeatureisanalyzedanddiscussed,andgaborwavelett
4、ransformationisusedtopickuptexturefeature.Afterwards,thispaperbrieflydescribesrelatedcontentofinvarianttheory,andchoosestheinvariantdescriptorofcombiningwithDFTandlogpolartransformationtobeimagecharacterdescriptor,anddoestherelevantimageretrievalexperiment.Relevancefeedbacktechniquehasb
5、eenanimportantapproachinimageretrievaldiscussedinthispaper.Meanwhilethispaperquotesanovelimageretrievalmethodofsupportvectormachinebasedrelevancefeedback,andtheproposedapproachpartiallysolvesthesampleinsufficiencyproblemthatexistsinSVMbasedrelevancefeedbackalgorithm,andshowtheapproachha
6、sgoodperformanceforretrieval.Theimageretrievalfusingwithmulti-characterisaprogressdirection.Howeverhowtoefficientlyfuseandaffirmtheweightvalueamongdifferentcharactersinthetraditionalrelevancefeedbackisabigproblem.Thispaper,withtheideaofinformationfusion,proposesatwo-layerSVMclassifiermo
7、deltoimprovetheprecisionoftheclassificationandusesSVM-basedInformationFusionMachinetoefficientlysolvetheproblemoffusingmulti-characterintheimageretrieval.Itradicallysolvesthebigproblemofdistillingandfusingmulti-character.Manyexperimentsprovethatthethreeimagecharacters,thatis,in
此文档下载收益归作者所有