欢迎来到天天文库
浏览记录
ID:36423267
大小:7.49 MB
页数:99页
时间:2019-05-10
《基于部分的人脸识别方法研究》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、中国科学技术大学博士学位论文基于部分的人脸识别方法研究姓名:潘可申请学位级别:博士专业:模式识别与智能系统指导教师:张培仁20090201摘要利用各部分的生物特征图像质量对该局部的识别结果的置信度进行评估:从而根据该置信度对各部分的识别结果在最终的结论中所占的比重进行动态调整。这就实现了尽可能的从变化程度低的人脸的局部获取正确的信息的同时,尽可能的减少从变化程度高的人脸局部对总体的结果的影响。实验表明该方法提高了系统的识别性能与对人脸形变、表情、姿态等的适应性,在相同的条件下,该方法超过了基于整体的方法5.6%,超过采用部分的固定加
2、权合规则的方法5%。。关键词:模式识别,人脸识别,图像质量评估,基于部分,统计学习,多数据融合·IIABSTRACTABSTRACTFacerecognitionisoneofthemostprominentareasincomputerVisionforseveraldecades.NumerousfacerecognitionmethodshaVebeenproposed.CurrentsystemscandofAirlyaccuraterecognjtionunderconstrainedscenariosusingthese
3、facerecog—nitiOnmethods.IthasattractedmuchattentiOnduetoitspotentialapplicationValues.Nowadays,facerecognitionhasgreatVaIueinpublicsafety,intelligencesur、,eillance,jdentifjcation,E—commerce,multimedia,digitalentertainmentandsoon.Requirementsofthepracticalityofthefacere
4、cognitionisincreasedalongwiththewjderuse.Also,mos£ofthecurrentfacerecogni£ionmethodscandofairlyaccuraterecognitionin£XedenVironment.Butthe6xedenvironmentconstrainisnotsuitableformostoftheapplication.Thispaperdiscusst11eproblemsmentionedabove,thefolIowingcontributionsar
5、emade:·Astatisticallearningbasedfacedecompositionmemodisproposed.InordertooVerComedif!fIerentsituationofhumanfaces,suchasshadingandemotion.Somescholarsproposedp2urt-basedorco玎1ponentbasedfacerecognitionmeth—ods.Butthedecompositionof量lumanfacesareallguidedundertheschola
6、rs’experiences.Thispaperproposedastatisticalleamingbasedfacedecomposi-tionmethods。Our∞ethoddecomposesthehumanfacemorescientifically,andreacbedahigherperfOrmance.Inthe1earningprocessOftheclassifiers,thede—compositionisalsoleamed.ThecontributionofeachpartisgiVen.Throught
7、heanalysis,thedecompositionisguidedundereachp狐’scontribution.Intllebasisoft}1edecompositionmethod,weproposedapart-basedfacerecognitionmethod.’I’heproposedme也odhasa4.53%perfo肿anceimprovementmanthe0riginalfacerecognitionmemod.··AnuniVersalbiometricimagequalityassessmentm
8、ethodisproposed.ThispapergiVesoutthede矗nitionofBiometricImageQualityandimagedistancefunctionandsoon.Throughthedefinit
此文档下载收益归作者所有