資源描述:
《基于視頻的人臉識別研究》由會員上傳分享,免費在線閱讀,更多相關(guān)內(nèi)容在學(xué)術(shù)論文-天天文庫。
1、華南理工大學(xué)博士學(xué)位論文基于視頻的人臉識別研究姓名:李盛陽申請學(xué)位級別:博士專業(yè):通信與信息系統(tǒng)指導(dǎo)教師:葉梧20041001摘要ABSRACTRecently,facerecognitioninvideoisdevelopedrapidlyduetotherequirementoftheapplicationofvideomonitor,human-computercommunicationandvideoretrieval.Humanfacerecognitioninvideoconsistsof
2、threeparts:humanfacedetectinginsingleframe,humanfacetracingofvideo,identificationofhumanfaceimagesequences.Thisdissertationisfocusedonthesethreeparts,andgainsachievementsinphases.Inaddition,thisdissertationalsoincludesaprimaryarithmeticevaluatingsystemo
3、fhumanfaceidentification,instrumentfingermonitortechnologyinindustrialprocessanddefaultdetectingtechnologybasedonminimalentropy.FacedetectionarithmeticbasedonDCTdomainisdevelopedinthisdissertation,becausemostvideoinformationexistsincompressingformat,and
4、DCTisgenerallyusedinvideoandpicturecompressing.ThisarithmeticaccomplishestheprimarydetectionofhumanfacebasedonGaussmodelinTSLcolorspaceatfirst,andthen,deletesnon-faceregionusingclassifierbasedonDCTgradientenergyandpreviousknowledge.Wedesigntwokindsofcla
5、ssifier,oneisusedinvalidatinghumanfaceregionbasedonvarianceofgradientenergy,theothersisbasedongradientenergytemplate.Experimentsshowthatthisarithmetichasbetterperformancesinsimplepersonsimplebackground,singlepersoncomplexbackground,multipersonsimpleback
6、groundandmultipersoncomplexbackground.BasedontheresearchofMPEGvideostream,facetracingsystemincomplexbackgroundbasedonMPEGstreamisproposed.Thesystemuseshistogramtodivideshot,andthenautomaticusesMPEGcodingcharacteristictotracehumanfacedetectedfromIframe.E
7、xperimentsshowthatthisarithmeticisrobustinshadingandturningsituations.Animprovedelasticgraphmatchingarithmeticisusedinidentificationofstatichumanfacefiguresandhumanfacesequencesinthisdissertation.Thisarithmeticlabelsfeaturepointbasedonenergydistributing
8、ofGaborwavelettransform,sonohumaninterventionisneededintrainingprocess.ExperimentsshowthatthisarithmetichashigheridentifyingratesthanEBGMarithmetic.Inordertoevaluatetheperformanceofthearithmeticofhumanfaceidentification,wedesigna