第37卷第3期2023年6月JournalofJiangsuUniversityofScienceandTechnology(NaturalScienceEdition)D0I:10.20061/j.issn.1673-4807.2023.03.012江苏科技大学学报(自然科学版)Vol.37No.3Jun.2023基于OEDLBP的人脸欺诈检测算法研究王艳,夏坤,束鑫(江苏科技大学计算机学院,镇江212100)摘要:人脸识别技术提高了身份验证的效率,给用户带来快捷方便的体验.但是,当有人试图伪造用户人脸通过系统验证时,就会威胁合法用户的信息和财产安全.针对打印攻击和视频攻击,提出了一种基于奇偶数位像素差异的描述子OEDLBP,该算法将局部偶数位差值模式(EDLBP)、局部奇数位差值模式(ODLBP)和全局特征模式(GBP)结合,利用空间金字塔算法统计彩色图像通道内和通道间特征,将提取到的特征进行融合并用SVM对真假人脸进行分类,在CASIA-FASD、Replay-Attack和Replay-Mobile3个人脸反欺骗数据库中取得了较好的实验效果。关键词:人脸欺诈检测;局部二值模式;奇偶位局部二值模式;空间金字塔;彩色空间中图分类号:TP391ResearchonfacespoofingdetectionalgorithmbasedonOEDLBP(SchoolofComputer,JiangsuUniversityofScienceandTechnology,Zhenjiang212100,China)Abstract:Facerecognitiontechnologyhasimprovedauthenticationefficiencyandgivenusersafastandconven-ientexperience.However,whensomeonetriestofaketheuser'sfacetopassthesystemverification,itwillthreatentheuser'sinformationandpropertysecurity.Thispaperproposesadescriptorbasedonoddandevenbitdifferencelocalbinarypattern(OEDLBP)forprintandvideoattacks.Thealgorithmcombinestheeven-bitdifferencelocalpattern(EDLBP),theodd-bitdifferencelocalpattern(ODLBP),andtheglobalfeaturepattern(GBP),andusesthespatialpyramidalgorithmtocounttheintra-channelandinter-channelfeaturesofthecolorimage.Then,theextractedfeaturesarefusedandthenclassifiedbySVM.Thisapproachachievesgoodperform-anceinthreechallengingfaceanti-spoofingdatabases:CASIAFASD,Replay-Attack,andReplay-Mobile.Keywords:facespoofingdetection,localbinarypattern,oddandevenbitlocalbinarypattern,spacepyramid,人脸识别技术已广泛应用于出人境管理、智能人脸图像中提取特定的特征;分类是指将未标记的安防、公安刑侦等领域,给用户带来便利的同时,其人脸图像分类为特定类别,以确定它们是活体人...