基于線性非對稱級聯(lián)分類器和膚色驗證組合的人臉檢測算法研究

1、南京理工大學(xué)碩士學(xué)位論文基于線性非對稱級聯(lián)分類器和膚色驗證組合的人臉檢測算法研究姓名：孫煒申請學(xué)位級別：碩士專業(yè)：控制理論與控制工程指導(dǎo)教師：郭健20080615Abstract碩士論文AbstractWi廿1thedevelopmentofsocietyhighspeedeffectivestatusauthenticationisneededurgentlyininteractivetechnology，accesscontrol，

2、criminaldetectiondonebyPolice，andidentityrecognitionetcComparedwithotherbiometriccharacteristicsrecognitionfacerecognitionisdirectconcealedconvenientfriendlysecure，andstable，itissmdiedbyalotofpeopleLinearasymmetriccascad

3、eclassifierbasedonfeaturefuzzyclassificationalgorithmandlinearasymmetriccascadeclassifierbasedonfeaturespace’SdoublethresholdalgorithmalestudiedinthispaperEachofthemiscombinedwithcolorverificationthentwealtimealgorithmsa

4、redesignedSoftwaresystemofthetwoalgorithmsisdeveloped11heworkisasfollows：Linear。asymmetriccascadeclassifierrealtimefacedetectionalgorithmbasedonfeaturefuzzyclassificationandit’Scombinationwithcolorverificationalgorithmar

5、estudiedinthispaperFirst，fuzzyclassificationtechnologyisusedinharrlikefeatureintraining，andharrlikefeatureweakclassifierisdesignedSecondFastFeatureSelection(FFS)algorithmisusedtoselectharrlikeweakclassificationforeveryno

6、deThird，LinearAsymmetricClassifier(LAC)algorithmisusedtodesignharr—likeweakclassificationselectedforeverynodeSOastoformstrongclassificationWKledetecting，anode’strongclassificationcanbechosentodetectfaceTemovenonskinandfi

7、ndoutface，thedetectingresultisdealtbycolorverificationExperimentsshowthatthealgorithmistimesavingintrainhighefficienldetectingandtimesavingindetection，andisofhishdetectionrateLinearasymmetriccascadeclassifierrealtimeface

8、detectionalgorithmbasedonfeaturespace’Sdual—thresholdandit’ScombinationwithcolorverificationalgorithmaresmdiedⅥ嘶letrainingharrlikefeatureculveoffaceandnonfacearedrawninthefeaturespaceandtwothresholdsarefindoutThenFastFea

9、tureSelectionalgorithmisusedtoselex：lhartlikeweakclassificationforeverynodeTtlird，LinearAsymmetricClassifierisusedtcdesignhartlikeweakclassificationselectedforeverynodeSO笛toformstrongclassificationLaSt’thesestrongclassif

10、icationsarecascadedtoformcascadeclassificationWhiledetectingpicturemustbedetectedbyeverynodeinturnThenthedetectingresultisdealtbycolo】verificationtoremovenonskinandfindoutfaceExperimentsshowthatthealgorithm“moreaccuratet

11、hantraditionalsinglethresholdandthetraintimeislessthanthemFacedetectionsoftwaresystemismadeinVisualCnet2003、】l，imOpenCVl(MachineVisionLibraryThesoftwareisoffriendlyinterface，convenientuse，freeinpulpictureformatandextensi