Vol.29 Suppl.2 袁立等：基于人脸和人耳的多模态生物特征识别 .193. 100 Vendor Test 2002:Evaluation Report.Technical Report.NIS- TIR 6965.Gaitherburg.National Institute of Standards and 95 Technology.2003.http/www.frvt.org/FRVT2002/docu- 90 ments.htm [2]Ross A,Jain A.Multimodal biometrics:an overview //Proc.of 85 ·一特征层加法融合 12th European Processing Conference.2004:1221 粒80 日一特征层乘法融合 +一图像层融合 [3]Yacoub S B.Abdeljaoued Y.Fusion of face and speech data for 75 人耳 person identity verification.IEEE Tran Neur Networks,1999. e-人脸 10(5):1065 700 30 4050607080 [4]Zhou X L.Bhanu B.Integrating face and gait for human recogni- 特征空间维数 tion//Proe.of Conference on Computer Vision and Pattern Recognition.New York.United States.2006:55 图3不同识别方法和融合方法的识别率对比 [5]Chang K.Bowyer K.Flynn P.Multi-biometrics using facial ap- pearance,shape and temperature//Proc.of the Sixth IEEE In- 上述实验结果表明基于人脸人耳信息融合的多 ternational Conference on Automatic Face and Gesture Recogni- 模态识别的识别率优于利用单体的人耳识别或人脸 tion:Seoul,South Korea,2004:43 识别，在融合识别中，采用加法规则的特征层融合 [6]Hong L,Jain A.Integrating faces and fingerprints for personal i- 的识别率高于采用乘法规则的特征层融合识别率和 dentification.IEEE Transactions on Pattern Analysis and Ma- 图像层的融合识别率，由于人脸图像库中图像质量 chine Intelligence,1998,20(12):1295 不如人耳图像，故识别率较低，本实验说明基于特 [7]Wang Y.Tan T,Jain A.Combining face and iris biometrics for identity verification//Proe.of 4th International Conference on 征融合的方法可以有效提高识别率. Audio Video Based Pattern Analysis.UK.2003:805 4结论 [8]Ross A.Jain A.Information fusion in biometrics.Pattern Recogn Lett.2003,(24):2115 本文提出一种新的将正面人脸和人耳进行融合 [9]Yuan L.Mu Z C.Zhang Y,et al.Ear recognition using im- 的多模态生物特征识别方法，在USTB人耳图像库 proved non negative matrix factorization /18th International Conference on Pattern Recognition.HK.2006 和ORL人脸图像库上，应用基于全空间的核Fisher [10]刘青山，卢汉清，马颂德.综述人脸识别中的子空间方法，自 鉴别分析方法进行识别，实验结果表明基于特征层 动化学报，2003,29(6)：900 和图像层融合策略的识别率优于利用单体特征的识 [11]陈才扣，杨静宇.基于组合子空间的最优特征抽取及人脸识 别率.下一步工作是扩建USTB人脸人耳多模态图 别.信号处理，2004,20(6)：609 像库，并在更大规模的图像库上进行实验，以对该系 [12]Yang J.Yang JY.Optimal FLD algorithm for facial feature ex- traction//SPIE Processing of the Intelligent Robots and Com 统进行性能评价 puter Vision XX:Algorithms.Techniques,and Active Vision. 2001,4572,438 参考文献 [1]Phillips P J.Grother P,Micheals R J.et al.Face Recognition Multimodal recognition using face and ear Y UAN Li,MU Zhichun,ZENG Hui Information Engineering of School.University of Science and Technology Beijing.Beijing 100083.China ABSTRACI Unimodal biometric systems have to contend with a variety of problems and sometimes cannot sat- isfy application requirements.In this paper,a novel method of multimodal recognition using frontal face and ear was proposed.Kernel Fisher Discriminant Analysis was used for ear recognition,face recognition and the multi- modal recognition.The multimodal recognition was studied on the image level fusion and feature level fusion. The experimental results from using USTB ear database and ORL face database show that the multimodal recog- nition outperforms the unimodal biometric recognition.This work shows that multibiometric system can increase the accuracy of overall system recognition,and provides an effective approach of non intrusive biometric recogni- tion. KEY WORDS ear recognition:face recognition;multimodal recognition;Kernel Fisher Discriminant Analysis图3 不同识别方法和融合方法的识别率对比 上述实验结果表明基于人脸人耳信息融合的多 模态识别的识别率优于利用单体的人耳识别或人脸 识别．在融合识别中‚采用加法规则的特征层融合 的识别率高于采用乘法规则的特征层融合识别率和 图像层的融合识别率．由于人脸图像库中图像质量 不如人耳图像‚故识别率较低．本实验说明基于特 征融合的方法可以有效提高识别率． 4 结论 本文提出一种新的将正面人脸和人耳进行融合 的多模态生物特征识别方法．在 USTB 人耳图像库 和 ORL 人脸图像库上‚应用基于全空间的核 Fisher 鉴别分析方法进行识别．实验结果表明基于特征层 和图像层融合策略的识别率优于利用单体特征的识 别率．下一步工作是扩建 USTB 人脸人耳多模态图 像库‚并在更大规模的图像库上进行实验‚以对该系 统进行性能评价． 参 考 文 献 ［1］ Phillips P J‚Grother P‚Micheals R J‚et al．Face Recognition Vendor Test 2002：Evaluation Report．Technical Report‚NIS￾TIR 6965‚Gaitherburg． National Institute of Standards and Technology‚2003．http ∥ www．frvt．org／FRVT2002／docu￾ments．htm ［2］ Ross A‚Jain A．Multimodal biometrics：an overview ∥Proc．of 12th European Processing Conference．2004：1221 ［3］ Yacoub S B‚Abdeljaoued Y．Fusion of face and speech data for person identity verification．IEEE Tran Neur Networks‚1999‚ 10（5）：1065 ［4］ Zhou X L‚Bhanu B．Integrating face and gait for human recogni￾tion∥ Proc．of Conference on Computer Vision and Pattern Recognition．New York‚United States‚2006：55 ［5］ Chang K‚Bowyer K‚Flynn P．Mult-i biometrics using facial ap￾pearance‚shape and temperature∥ Proc．of the Sixth IEEE In￾ternational Conference on Automatic Face and Gesture Recogni￾tion‚Seoul‚South Korea‚2004：43 ［6］ Hong L‚Jain A．Integrating faces and fingerprints for personal i￾dentification．IEEE Transactions on Pattern Analysis and Ma￾chine Intelligence‚1998‚20（12）：1295 ［7］ Wang Y‚Tan T‚Jain A．Combining face and iris biometrics for identity verification∥ Proc．of 4th International Conference on Audio Video Based Pattern Analysis．UK‚2003：805 ［8］ Ross A‚Jain A．Information fusion in biometrics．Pattern Recogn Lett‚2003‚（24）：2115 ［9］ Yuan L‚Mu Z C‚Zhang Y‚et al．Ear recognition using im￾proved non-negative matrix factorization ∥ 18th International Conference on Pattern Recognition．HK‚2006 ［10］ 刘青山‚卢汉清‚马颂德．综述人脸识别中的子空间方法．自 动化学报‚2003‚29（6）：900 ［11］ 陈才扣‚杨静宇．基于组合子空间的最优特征抽取及人脸识 别．信号处理‚2004‚20（6）：609 ［12］ Yang J‚Yang J Y．Optimal FLD algorithm for facial feature ex￾traction∥ SPIE Processing of the Intelligent Robots and Com￾puter Vision XX：Algorithms‚Techniques‚and Active Vision‚ 2001‚4572：438 Multimodal recognition using face and ear Y UA N L i‚MU Zhichun‚ZENG Hui Information Engineering of School‚University of Science and Technology Beijing‚Beijing100083‚China ABSTRACT Unimodal biometric systems have to contend with a variety of problems and sometimes cannot sat￾isfy application requirements．In this paper‚a novel method of multimodal recognition using frontal face and ear was proposed．Kernel Fisher Discriminant Analysis was used for ear recognition‚face recognition and the multi￾modal recognition．The multimodal recognition was studied on the image level fusion and feature level fusion． The experimental results from using USTB ear database and ORL face database show that the multimodal recog￾nition outperforms the unimodal biometric recognition．This work shows that multibiometric system can increase the accuracy of overall system recognition‚and provides an effective approach of non-intrusive biometric recogni￾tion． KEY WORDS ear recognition；face recognition；multimodal recognition；Kernel Fisher Discriminant Analysis Vol．29Suppl．2 袁 立等： 基于人脸和人耳的多模态生物特征识别 ·193·