基于超特征的人脸识别
|
摘要
人脸识别技术是当前模式识别和图像处理领域的热点和难点,属于生物特征识别技术中的一种。生物特征识别技术是利用生物体(一般特指人)本身的生物特征来对生物体个体进行区分的方法,具有广泛的应用前景,所以开展人脸识别的研究具有重大意义。
传统的人脸识别方法需要大量的训练样本,而在实际的应用中,我们往往无法获得足够数量的样本,甚至可能在只有一幅某人的人脸图像的情况下,就要判断该图像中所记录的人脸的真实身份,在这种情况下,传统的方法的识别率会显著下降,甚至会无法达到实际应用的要求。
基于超特征的物体身份识别方法可以实现在只有少数,甚至是一个训练图像的前提下,依据由模块的多个特征组成的超特征量,自动地估算出该图像中信息量较大的块,并依据概率统计特性,将所有标定出的图像块按其所提供的信息量由大到小进行排序,取前面信息量大的M块进行识别判断。
另一方面,人脸是非刚体,存在表情变化,且人脸随年龄的增长而有所改变;此外,发型、眼镜对人脸造成遮挡,且人脸成像容易受光照、成像角度、成像距离等影响,这些都是人脸识别领域有待解决的难题。而基于超特征的识别法采用了分块的机制,且将主要的判决侯选块划定在人的眼睛和嘴巴周围,这样就减少了由于表情变化和发型的改变等引起的脸型和人脸图像中五官间相对位置的变化,以及遮挡等因素带来的影响。
本文对超特征身份识别法和经典的PCA人脸识别法进行了学习和研究,结合人脸识别的具体应用,设计了基于超特征的人脸识别系统和基于PCA的人脸识别系统。并针对经过检测且进行了各种归一化处理后的人脸图像,就评价人脸识别系统的几个重要参数,如:正确识别率、训练样本库所消耗的时间、识别一幅人脸图像所需时间等分别对两个系统进行了测试。通过对ORL人脸数据库和实验室采集的人脸库进行测试的识别结果进行比较和分析,实验结果说明了基于超特征的身份识别法应用于人脸识别系统可以有效地解决人脸图像识别中高维度、小样本的难题,在一定程度上克服了传统的人脸识别方法对于训练样本数和样本标准性的诸多要求,具有很强的学习和推广性。
As a focus and difficulty in the field of Pattern Recognition and Image Processing, Face Recognition(FR),which Belongs to the biometrical identification technology, distinguishing between individual instances by their biologic feature,has an important significance and widely practical application.
Traditional methods of Face Recognition require a number of training samples. However,in the practical,we are usually unable to obtain sufficient number of samples. And may even under the condition of having only one facial image,we must identity who is it.In this case,recognition rate of traditional methods declined significantly,and even unable to achieve the need of practical application.
Under the condition of few or even only one instance,object identification based on hyper-feature can estimates the informative patches automatically,according to the vectors of hyper-feature which includes many features selected from the patch.In additional, according to the Statistical Learning Theory,it can also make a decision by matching the number of M most informative patches.
On the other hand,the appearance changes with age and expression.In addition, hairstyle,glasses on the face and the light,imaging angle,distance and so on will easily affect the recognize result.Object identification based on hyper-feature,which adopts a patch-based mechanism,selecting the candidate patches around eyes and mouth,will reduce the affect caused by the change to a certain extent.
This paper mainly on studying and researching the algorithms of object identification based on hyper-feature and PCA.With the application in face recognition,design the face recognition system based on hyper-feature and PCA respectively.In order to test the systems,we design testing parameter as recognition rate,time-consuming in training and recognizing one image,using the images after face detection and some pretreatment as input.By testing ORL faces and laboratory database,as well as analyzing the data of the experiment,the result show that the application of object identification based on hyper-feature in face recognition could solute the problem of the high dimension and small sample in the field of face recognition to a certain extent.In addition,the method not only reduces the limit of traditional one,such as the number of training data,the criterion of instances and so on,but also has a widely practical application and important signification.
传统的人脸识别方法需要大量的训练样本,而在实际的应用中,我们往往无法获得足够数量的样本,甚至可能在只有一幅某人的人脸图像的情况下,就要判断该图像中所记录的人脸的真实身份,在这种情况下,传统的方法的识别率会显著下降,甚至会无法达到实际应用的要求。
基于超特征的物体身份识别方法可以实现在只有少数,甚至是一个训练图像的前提下,依据由模块的多个特征组成的超特征量,自动地估算出该图像中信息量较大的块,并依据概率统计特性,将所有标定出的图像块按其所提供的信息量由大到小进行排序,取前面信息量大的M块进行识别判断。
另一方面,人脸是非刚体,存在表情变化,且人脸随年龄的增长而有所改变;此外,发型、眼镜对人脸造成遮挡,且人脸成像容易受光照、成像角度、成像距离等影响,这些都是人脸识别领域有待解决的难题。而基于超特征的识别法采用了分块的机制,且将主要的判决侯选块划定在人的眼睛和嘴巴周围,这样就减少了由于表情变化和发型的改变等引起的脸型和人脸图像中五官间相对位置的变化,以及遮挡等因素带来的影响。
本文对超特征身份识别法和经典的PCA人脸识别法进行了学习和研究,结合人脸识别的具体应用,设计了基于超特征的人脸识别系统和基于PCA的人脸识别系统。并针对经过检测且进行了各种归一化处理后的人脸图像,就评价人脸识别系统的几个重要参数,如:正确识别率、训练样本库所消耗的时间、识别一幅人脸图像所需时间等分别对两个系统进行了测试。通过对ORL人脸数据库和实验室采集的人脸库进行测试的识别结果进行比较和分析,实验结果说明了基于超特征的身份识别法应用于人脸识别系统可以有效地解决人脸图像识别中高维度、小样本的难题,在一定程度上克服了传统的人脸识别方法对于训练样本数和样本标准性的诸多要求,具有很强的学习和推广性。
As a focus and difficulty in the field of Pattern Recognition and Image Processing, Face Recognition(FR),which Belongs to the biometrical identification technology, distinguishing between individual instances by their biologic feature,has an important significance and widely practical application.
Traditional methods of Face Recognition require a number of training samples. However,in the practical,we are usually unable to obtain sufficient number of samples. And may even under the condition of having only one facial image,we must identity who is it.In this case,recognition rate of traditional methods declined significantly,and even unable to achieve the need of practical application.
Under the condition of few or even only one instance,object identification based on hyper-feature can estimates the informative patches automatically,according to the vectors of hyper-feature which includes many features selected from the patch.In additional, according to the Statistical Learning Theory,it can also make a decision by matching the number of M most informative patches.
On the other hand,the appearance changes with age and expression.In addition, hairstyle,glasses on the face and the light,imaging angle,distance and so on will easily affect the recognize result.Object identification based on hyper-feature,which adopts a patch-based mechanism,selecting the candidate patches around eyes and mouth,will reduce the affect caused by the change to a certain extent.
This paper mainly on studying and researching the algorithms of object identification based on hyper-feature and PCA.With the application in face recognition,design the face recognition system based on hyper-feature and PCA respectively.In order to test the systems,we design testing parameter as recognition rate,time-consuming in training and recognizing one image,using the images after face detection and some pretreatment as input.By testing ORL faces and laboratory database,as well as analyzing the data of the experiment,the result show that the application of object identification based on hyper-feature in face recognition could solute the problem of the high dimension and small sample in the field of face recognition to a certain extent.In addition,the method not only reduces the limit of traditional one,such as the number of training data,the criterion of instances and so on,but also has a widely practical application and important signification.
引文
[1]肖冰,王映辉,人脸识别研究综述,计算机应用研究,2005,8,1-5.
[2]周杰,卢春雨等,人脸自动识别方法综述,电子学报,2000,28(4),102-106.
[3]Samal A,Iyengar P A.Automatic Recognition and Analysis of Human Faces and Facial Expressions:A Survey.Pattern Recognition,1992,25(1):65-77
[4]李月敏,陈杰,高文,尹宝才,快速人脸检测技术综述
[5]周激流,张晔,人脸识别理论研究进展,计算机辅助设计与图形学学报,1999,11(2):180-184
[6]艾英山,张德贤,人脸识别方法的综述与展望,计算机与数字工程,2005,33(10),24-27
[7]苏剑波,徐波,应用模式识别技术导论—人脸识别与语音识别。上海:上海交通大学出版社,2001
[8]R.Chellappa,C.L.Wilson,S.Sirohey and C.S.Barnes,Human andmachine recognition of faces:A survey.Proceedings of the IEEE,1995,83(5):705-740.
[9]周杰,卢春雨,张长水,李衍达,人脸自动识别方法综述,电子学报,2000,28(4):102-106
[10]梁路宏,艾海舟,徐光祜等,人脸检测研究综述,计算机学报,2002,VOL.25,NO.5,pp 449-459
[11]S.A.Sirohey,Human Face Segmentation and Identification,Technical Report,CS-TR-3176,Univ.of Maryland,1993.
[12]Govindaraju V,Srihari S.N,Sher D.B.A computational model for face location.In:Proc.IEEE Conf.on Computer Vision,Osaka,Japan,1990,718-721
[13]A.Gammerman,V.Vapnik,and V.Vowk,Learning by transduction,in Conference of Uncertainty in Artificaial Intelligence,pp.148-156.
[14]施善昌,自动识别技术原理及应用[M]。北京:人发邮电出版社,1999.
[15]周志明,王以治,黄文芝,王宁宁,基于小波的人脸识别技术,计算机工程与应用,2004,12,52-54。
[16]Lifang Wu,Lansun Shen,Kebin Jia etc.Improvement of Automatic Human Face Detection[S].ISO/IEC JTC1/SC29/WG11 MPEG2000/m6536,La Baule,Frarice, M.Pickering,A.Whichello,M.Frater etc.A proposal for an Automatic face Extraction Algorithm[S].ISO/IEC JTC1/SC29/WG11 MPEG2000/m6536,La Baule,France(The two Algorithms are combined in m6536),2000.
[17]H.Martin,H.Hunke.Locating and Tracking of Human Faces with Neural Networks.TechnicalReport of CMU,CMU-Cs-94-155,1994.
[18]M.Turk,A.Pentland.Eigenfaces for recognition.J.Cog.Neurosci.3,pp.71-86,1991.
[19]Sung K.Poggio T.Example-based learning for view based human face recognition.IEEE Trans.-On PAMI,1998,20(1),pp.39-51.
[20]Garcia C,Zikos G,Tziritas G,Face detection in color images using wavelet packet analysis,In:Proc.Multimedia Computing and Systems,Centro Affair,Florence,Italy,1999,VOL.1,pp.703-708.
[21]M.Propp,A.Samal,Artificial neural network architecture for human face recognition,Intell.Eng.Systems Artificial Neural Networks 2,1992,pp.535-540.
[22]T.Agui,Y.Kokubo,H.Nagashashi,etc.Extraction of face recognition from monochromatic photographs using neural networks.In Proceedings of the Second International Conference on Automation,Robotics and computer Vision,1992,VOL.1,pp.CV 18.8.1-CV 18.8.5.
[23]H.A.Rowley,S.Baluja,T.Kanade,Neural network-based face recognition,IEEE Trans.Pattern Anal.Mach.Intell.20,January 1998,pp.23-38.
[24]V.Vapnik,The Nature of Statistical Learning Theory.New York:Springer-Verlag,1995.
[25]程云鹏,矩阵论,第二版,西北工业大学出版社,2004.
[26]苏宏涛,基于统计特征的人脸识别技术研究,[博士学位论文],西安,西北工业大学出版社,2005.
[27]H.Schneiderman,T.Kanade,Probabilistic modeling of local appearance and spatial relationships for object recognition.In:Proc.IEEE Conf.on Computer Vision and Pattern Recognition,Santa Barbara,California,1998,pp.45-51.
[28]A.V.Nefian,M.H.Hayes,Face detection and recognition using Eigenface models.In:Proc.IEEE Conf.on Image Processing,Chicago,1998,pp.141-145.
[29]A.V.Nefian,M.H.Hayes,An embedded HMM- based approach for face detection and recognition.In:Proc.IEEE Conf.on Acoustics,Speech,and Signal Processing,Phoenix,Arizona,1999,VOL.6,pp.3553-3556.
[30]Paul Viola,Michael Jones.Rapid Object Detection using a Boosted Cascade of Simple Features.IEEE Computer Vision and Pattern Recognition(CVPR' 01),VOL.1,2001.
[31]S.Z.Li,L.Zhu,Z.Q.Zhang etc.Statistical Learning of Multi-view Face Recognition.In:ECCV 2002,pp.67-81
[32]Richard O.Duda,Peter E.Hart,David G.Stork.模式分类,北京:机械工业出版社.2003.
[33]T.KANADE.Picture Processing by computer complex and recognition of human faces[D],Dept.Inform.Sci,Univ of Kyoto,Tech,Rep,1993.
[34]Constantine P.Papageorgiou,Michael Oren,Tomaso Poggio.A General Framework for Object Identification.Proceedings of International Conference on Computer Vision.January 1998.
[35]Mohan,C.Papageorgiou,T.Poggio.Example-based object identification in images by components.IEEE Transactions on Pattern Analysis and Machine Intelligence,2001,VOL.23,NO.4,pp.349-361.
[36]高文,陈熙霖,计算机视觉——算法与系统原理,清华大学出版社,2002.
[37]Lienhart R,Kuranov A,V Pisarevsky.Empirical analysis of detection cascades of boosted classifiers for rapid Object Identification.DAGM' 03 25th Pattern Recognition Symposium 2003.
[38]F.Crow.Summed-area tables for texture mapping.Proceedings of SIGGGRAPH,1984,VOL.18(3),pp.207-212.
[39]Philippe G.Lacoute.Fast Volume Rendering Using a Shear-Warp Factorizaing of the ViewingTransformation,Stanford University:Software Publishing Corp.1995,pp.68-83.
[40]Vapnik V N.Estimation of Dependencies based on Empirical Data.Berlin: Springer-Verlag,1997.
[41]Erik Learned-Miller,Jitendra Malik,etc.Building a Classification Cascade for Visual Identification from One Example.[J].Pattern Recognition,2006,32(7),pp.1237-1248.
[42]Andras Ferencz,Erik G.Learned-Miller,Jitendra Malik.Learning to Locate Informative Features for Visual Identi cation.www.eecs.berkeley.edu/Research/Projects/CS/vision/shape/vid
[43]边肇祺 张学工 等,模式识别,北京:清华大学出版社,2000
[44]Breiman L.Bagging Predictors.Machine Learning,1996,24(2):123-140
[2]周杰,卢春雨等,人脸自动识别方法综述,电子学报,2000,28(4),102-106.
[3]Samal A,Iyengar P A.Automatic Recognition and Analysis of Human Faces and Facial Expressions:A Survey.Pattern Recognition,1992,25(1):65-77
[4]李月敏,陈杰,高文,尹宝才,快速人脸检测技术综述
[5]周激流,张晔,人脸识别理论研究进展,计算机辅助设计与图形学学报,1999,11(2):180-184
[6]艾英山,张德贤,人脸识别方法的综述与展望,计算机与数字工程,2005,33(10),24-27
[7]苏剑波,徐波,应用模式识别技术导论—人脸识别与语音识别。上海:上海交通大学出版社,2001
[8]R.Chellappa,C.L.Wilson,S.Sirohey and C.S.Barnes,Human andmachine recognition of faces:A survey.Proceedings of the IEEE,1995,83(5):705-740.
[9]周杰,卢春雨,张长水,李衍达,人脸自动识别方法综述,电子学报,2000,28(4):102-106
[10]梁路宏,艾海舟,徐光祜等,人脸检测研究综述,计算机学报,2002,VOL.25,NO.5,pp 449-459
[11]S.A.Sirohey,Human Face Segmentation and Identification,Technical Report,CS-TR-3176,Univ.of Maryland,1993.
[12]Govindaraju V,Srihari S.N,Sher D.B.A computational model for face location.In:Proc.IEEE Conf.on Computer Vision,Osaka,Japan,1990,718-721
[13]A.Gammerman,V.Vapnik,and V.Vowk,Learning by transduction,in Conference of Uncertainty in Artificaial Intelligence,pp.148-156.
[14]施善昌,自动识别技术原理及应用[M]。北京:人发邮电出版社,1999.
[15]周志明,王以治,黄文芝,王宁宁,基于小波的人脸识别技术,计算机工程与应用,2004,12,52-54。
[16]Lifang Wu,Lansun Shen,Kebin Jia etc.Improvement of Automatic Human Face Detection[S].ISO/IEC JTC1/SC29/WG11 MPEG2000/m6536,La Baule,Frarice, M.Pickering,A.Whichello,M.Frater etc.A proposal for an Automatic face Extraction Algorithm[S].ISO/IEC JTC1/SC29/WG11 MPEG2000/m6536,La Baule,France(The two Algorithms are combined in m6536),2000.
[17]H.Martin,H.Hunke.Locating and Tracking of Human Faces with Neural Networks.TechnicalReport of CMU,CMU-Cs-94-155,1994.
[18]M.Turk,A.Pentland.Eigenfaces for recognition.J.Cog.Neurosci.3,pp.71-86,1991.
[19]Sung K.Poggio T.Example-based learning for view based human face recognition.IEEE Trans.-On PAMI,1998,20(1),pp.39-51.
[20]Garcia C,Zikos G,Tziritas G,Face detection in color images using wavelet packet analysis,In:Proc.Multimedia Computing and Systems,Centro Affair,Florence,Italy,1999,VOL.1,pp.703-708.
[21]M.Propp,A.Samal,Artificial neural network architecture for human face recognition,Intell.Eng.Systems Artificial Neural Networks 2,1992,pp.535-540.
[22]T.Agui,Y.Kokubo,H.Nagashashi,etc.Extraction of face recognition from monochromatic photographs using neural networks.In Proceedings of the Second International Conference on Automation,Robotics and computer Vision,1992,VOL.1,pp.CV 18.8.1-CV 18.8.5.
[23]H.A.Rowley,S.Baluja,T.Kanade,Neural network-based face recognition,IEEE Trans.Pattern Anal.Mach.Intell.20,January 1998,pp.23-38.
[24]V.Vapnik,The Nature of Statistical Learning Theory.New York:Springer-Verlag,1995.
[25]程云鹏,矩阵论,第二版,西北工业大学出版社,2004.
[26]苏宏涛,基于统计特征的人脸识别技术研究,[博士学位论文],西安,西北工业大学出版社,2005.
[27]H.Schneiderman,T.Kanade,Probabilistic modeling of local appearance and spatial relationships for object recognition.In:Proc.IEEE Conf.on Computer Vision and Pattern Recognition,Santa Barbara,California,1998,pp.45-51.
[28]A.V.Nefian,M.H.Hayes,Face detection and recognition using Eigenface models.In:Proc.IEEE Conf.on Image Processing,Chicago,1998,pp.141-145.
[29]A.V.Nefian,M.H.Hayes,An embedded HMM- based approach for face detection and recognition.In:Proc.IEEE Conf.on Acoustics,Speech,and Signal Processing,Phoenix,Arizona,1999,VOL.6,pp.3553-3556.
[30]Paul Viola,Michael Jones.Rapid Object Detection using a Boosted Cascade of Simple Features.IEEE Computer Vision and Pattern Recognition(CVPR' 01),VOL.1,2001.
[31]S.Z.Li,L.Zhu,Z.Q.Zhang etc.Statistical Learning of Multi-view Face Recognition.In:ECCV 2002,pp.67-81
[32]Richard O.Duda,Peter E.Hart,David G.Stork.模式分类,北京:机械工业出版社.2003.
[33]T.KANADE.Picture Processing by computer complex and recognition of human faces[D],Dept.Inform.Sci,Univ of Kyoto,Tech,Rep,1993.
[34]Constantine P.Papageorgiou,Michael Oren,Tomaso Poggio.A General Framework for Object Identification.Proceedings of International Conference on Computer Vision.January 1998.
[35]Mohan,C.Papageorgiou,T.Poggio.Example-based object identification in images by components.IEEE Transactions on Pattern Analysis and Machine Intelligence,2001,VOL.23,NO.4,pp.349-361.
[36]高文,陈熙霖,计算机视觉——算法与系统原理,清华大学出版社,2002.
[37]Lienhart R,Kuranov A,V Pisarevsky.Empirical analysis of detection cascades of boosted classifiers for rapid Object Identification.DAGM' 03 25th Pattern Recognition Symposium 2003.
[38]F.Crow.Summed-area tables for texture mapping.Proceedings of SIGGGRAPH,1984,VOL.18(3),pp.207-212.
[39]Philippe G.Lacoute.Fast Volume Rendering Using a Shear-Warp Factorizaing of the ViewingTransformation,Stanford University:Software Publishing Corp.1995,pp.68-83.
[40]Vapnik V N.Estimation of Dependencies based on Empirical Data.Berlin: Springer-Verlag,1997.
[41]Erik Learned-Miller,Jitendra Malik,etc.Building a Classification Cascade for Visual Identification from One Example.[J].Pattern Recognition,2006,32(7),pp.1237-1248.
[42]Andras Ferencz,Erik G.Learned-Miller,Jitendra Malik.Learning to Locate Informative Features for Visual Identi cation.www.eecs.berkeley.edu/Research/Projects/CS/vision/shape/vid
[43]边肇祺 张学工 等,模式识别,北京:清华大学出版社,2000
[44]Breiman L.Bagging Predictors.Machine Learning,1996,24(2):123-140