欢迎来到天天文库
浏览记录
ID:36807504
大小:2.31 MB
页数:74页
时间:2019-05-15
《基于支持向量机的人脸检测的研究》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、Abs仃actAbstractWi也thedevelopmentofthecomputerandinformationtechnology,imageprocessinggetsmuchmorewidelyusedinmanyfields,suchasscience,nationaldefense,industryandcommerce,finance,ere.Manyrecentworkswereinvolvedinhumanfacebecauseitcouldbecomeafriendlymutualinterfaceandalsocouldpro
2、videagreatdealofvaluableinformation.Facedetectionisanimportanttaskinhumanfaceinformationprocessing.Ithasbecomeahotsubjectinthefieldsofpattemrecognitionandcomputervisioninrecentyears.Nowadays,besidesfacerecognition,facedetectionisalsowidelyappliedinmanypromisingareassuchasvideosu
3、pervision,identityvalidation,multimediadatabaseretrieval,content—basedcompressionandretrievalinnetworktransmission,etc,Thispaperisconcemedaboutthekeyproblemsinimageclassificationsuchasfacedetection.Asetofmethodsandtheorieswereintroduced,Weintroducedanewvaliditymethodoffeatureext
4、ractionbasedondiscriminatingfeatureanalysis.Thediscriminatingfeatureanalysisderivesanewfeaturevectorforfacedetection,bycombiningtheinputimage,its1DHarrwaveletrepresentation,anditsamplitudeprojections.Whilegettingtheintegratedinformationoftheimages.theamplitudeprojectionsareenhan
5、cedwitlldiscriminatingpowerforhumanfacesbecauseoftheabilitytocapturetheverticalsymmetricdistributionsandthehorizontalcharacteristicsofhumanfaceimages.Theexperimentalresultshowsthismethodcalldescribethefeaturesoftheoriginalimageswithvalidlygoodcomputationalefficiency。Ithasbeensho
6、wneffectiveforhumanfaceandpedestriandetection.WedevelopedafaceimagedetecfionsystembasedonSVMclassifier.ThetheoryofSupportVectorMachineswasproposedundertheStatisticalLearningtheorywhichbasedonStructuralRiskMinimizationPrinciple.ThemethodofSVMgetsbeuergeneralizationthanArtificialN
7、euralNetworks(ANN),whichbasedonEmpiricalRiskMinimizationPrinciple.AnditwasprovedtobeapplicableinAbstractcomplicatedfacedetectionproblems.WedevelopedafaceimagedetectionsystembasedonSVMclassifiertosolvetheproblemoffrontalfacedetectionundercomplexbackgrounds.Inordertooptimizethesea
8、rchingregionsandimprovedetectionrateforthedetec
此文档下载收益归作者所有