一种有监督双向特征融合的人脸识别算法
|
摘要
人脸识别因其直接、友好、非侵犯性等特点成为当前生物特征识的焦点。但是人脸图像维数很高,并且需要较大的空间进行信息存储,因此人脸特征提取就显得非常重要。特征提取就是抽取人脸图像本身具有最大鉴别能力的特征,利用提取到的鉴别特征进行识别。本文对特征提取算法中的典型LPP算法及其改进算法进行了研究。LPP算法是一种无监督的针对1维向量的特征提取方法,转化过程中会出现“小样本”和维数过高的问题。SLPP算法和2DLPP算法是LPP的改进算法,SLPP算法解决了LPP算法的无监督问题,2DLPP算法可以针对2维人脸图像直接处理,避免了图像转化过程中的问题。但是SLPP算法中k近邻图构造时存在类内和类间两个k值难以确定的问题,2DLPP算法存在不能完整保留人脸整体特征的问题。针对两种算法的上述问题分别进行了改进,并把改进后的两种算法相结合提出一种有监督双向特征融合人脸识别算法,在标准人脸库进行了实验,通过实验对比表明本文提出的算法与其他人脸识别算法相比具有更高人脸识别率和鲁棒性。
本文对LPP及其改进算法进行了研究,主要贡献概括为以下三个方面:
第一,在SLPP算法的基础上针对其构造k近邻图时存在类内和类间两个k值难以确定问题进行了改进,对类间k近邻图构造时k的取值进行了修改,解决了类内和类间k进邻图构造时k值的确定问题,并把改进后算法与原有SLPP算法进行了实验对比,进而证明了SLPP算法修改的合理性。
第二,在传统单向2DLPP算法的基础上提出了双向特征融合算法—2DDLPP算法,将同一原始人脸图像映射到水平和垂直两个不同的特征空间中,得到互补的两类人脸图像特征,并对两类人脸特征进行融合,很好的保留了人脸整体特征和判别信息,并对改进后的2DDLPP算法进行了实验,验证了其改进的合理性。
第三,把有监督与2维双向图像特征提取相结合对LPP算法进行改进,提出一种新的有监督双向特征融合人脸识别算法—2DDESLPP算法,解决了LPP算法的“小样本问题”和无监督问题,通过实验对比证明该方法具有较高的识别率和鲁棒性。
Face identification is considered as the focus of biometric identification because of its characteristics of being direct, friendly and non-invasive. However, the high dimensionality of face images, and the need for more space for information storage lead to the importance of facial feature extraction. Feature extraction refers to the extraction of the most distinctive face images which are used for identification. This paper is a research on the typical LPP algorithm of feature extraction algorithm and its improved version. LPP is an unsupervised algorithm for 1-dimensional vector feature extraction, which may have the problem of "small sample" and excessively high dimension in the process of. SLPP algorithm and 2DLPP algorithm are two improved versions of LPP algorithm. SLPP algorithm solved the unsupervised problem of LPP algorithm and 2DLPP algorithm for the direct processing of 2-dimensional face image avoids the problem that may occur in the transformation process. But there are problems of SLPP algorithm on constructing the neighbor graph of within-class and between-class. 2DLPP algorithm has the limitation of being unable to retain the overall face feature. The paper try to make an improvement in the above two algorithms respectively, then put forward a new supervised two-way fusion face recognition algorithm that combines the two algorithms. Compared with other face recognition algorithm, the new algorithm is proved to be of higher recognition rate and robustness after being tested by experiments in the standard face database.
This paper studies LPP and two improved versions of it, of which the major contributions can be summarized in the following three aspects.
First, there are problems of SLPP algorithm on constructing the neighbor graph of within-class and between-class. We propose a new method to construct the graph between-class, which solve the above problem. We prove the consistency of the proposed ESLPP by number of experiments, and then prove that a SLPP algorithm change is reasonable.
Secondly, it put forwards the two-way feature fusion algorithm-2DDLPP based on the traditional 2DLPP algorithm, in which the primitive face images are mapped to the horizontal and vertical, two different feature spaces to get two complementary features of face images. The fusion of these two types of facial features well retains the overall characteristics of the human face and distinctive information. More importantly the improved 2DDLPP algorithm is acknowledged to be valid through experiment.
Thirdly, the newly proposed 2DDESLPP algorithm solves the problem of "small sample" and excessively high dimension and it is proved that the method has higher recognition rate and robustness.
本文对LPP及其改进算法进行了研究,主要贡献概括为以下三个方面:
第一,在SLPP算法的基础上针对其构造k近邻图时存在类内和类间两个k值难以确定问题进行了改进,对类间k近邻图构造时k的取值进行了修改,解决了类内和类间k进邻图构造时k值的确定问题,并把改进后算法与原有SLPP算法进行了实验对比,进而证明了SLPP算法修改的合理性。
第二,在传统单向2DLPP算法的基础上提出了双向特征融合算法—2DDLPP算法,将同一原始人脸图像映射到水平和垂直两个不同的特征空间中,得到互补的两类人脸图像特征,并对两类人脸特征进行融合,很好的保留了人脸整体特征和判别信息,并对改进后的2DDLPP算法进行了实验,验证了其改进的合理性。
第三,把有监督与2维双向图像特征提取相结合对LPP算法进行改进,提出一种新的有监督双向特征融合人脸识别算法—2DDESLPP算法,解决了LPP算法的“小样本问题”和无监督问题,通过实验对比证明该方法具有较高的识别率和鲁棒性。
Face identification is considered as the focus of biometric identification because of its characteristics of being direct, friendly and non-invasive. However, the high dimensionality of face images, and the need for more space for information storage lead to the importance of facial feature extraction. Feature extraction refers to the extraction of the most distinctive face images which are used for identification. This paper is a research on the typical LPP algorithm of feature extraction algorithm and its improved version. LPP is an unsupervised algorithm for 1-dimensional vector feature extraction, which may have the problem of "small sample" and excessively high dimension in the process of. SLPP algorithm and 2DLPP algorithm are two improved versions of LPP algorithm. SLPP algorithm solved the unsupervised problem of LPP algorithm and 2DLPP algorithm for the direct processing of 2-dimensional face image avoids the problem that may occur in the transformation process. But there are problems of SLPP algorithm on constructing the neighbor graph of within-class and between-class. 2DLPP algorithm has the limitation of being unable to retain the overall face feature. The paper try to make an improvement in the above two algorithms respectively, then put forward a new supervised two-way fusion face recognition algorithm that combines the two algorithms. Compared with other face recognition algorithm, the new algorithm is proved to be of higher recognition rate and robustness after being tested by experiments in the standard face database.
This paper studies LPP and two improved versions of it, of which the major contributions can be summarized in the following three aspects.
First, there are problems of SLPP algorithm on constructing the neighbor graph of within-class and between-class. We propose a new method to construct the graph between-class, which solve the above problem. We prove the consistency of the proposed ESLPP by number of experiments, and then prove that a SLPP algorithm change is reasonable.
Secondly, it put forwards the two-way feature fusion algorithm-2DDLPP based on the traditional 2DLPP algorithm, in which the primitive face images are mapped to the horizontal and vertical, two different feature spaces to get two complementary features of face images. The fusion of these two types of facial features well retains the overall characteristics of the human face and distinctive information. More importantly the improved 2DDLPP algorithm is acknowledged to be valid through experiment.
Thirdly, the newly proposed 2DDESLPP algorithm solves the problem of "small sample" and excessively high dimension and it is proved that the method has higher recognition rate and robustness.
引文
[1]裴佳佳.基于子空间的人脸识别技术研究[D].浙江工业大学2009.
[2] Shakhnarovich G, Moghaddam B. Face Recognition in Subspaces [M]. New York: Sp ringer-Verlag, 2004.
[3] TURK M,PENTLAND A P. Face recognition using Eigen faces [C].IEEE Conference on Computer Vision and Pattern Recognition. Los Akanitos, 1991:586-591.
[4] ZHUANG Z M, ZHANG A N, LI F L. Based on an optimized LDA algorithm for face recognition [J]. Journal of Electronics & Information Technology, 2007, 29(9):2047-2049.
[5] Zhou DL, Zhao DB. Face recognition based on singular value decomposition and discriminant KL projection.In: Journal of Software, 2003, 14(4):783-789.
[6] AH, and E Oja. Independent component analysis: algorithms and applications. In: Neural Networks, 2000, 13(4):411-430.
[7] Pcnio S Pencv. Local feature analysis:a statistical theory for information representation and transmission.New York:Rockefeller University,1998.
[8] Borg I and Grocnen P Modern.Multidimensional Scaling.In:Theory and Applications, New York:Springer-Verlag, 1997.
[9] YANG J, ZHANG D, and FRANGI A.F, et al. Two-dimensional PCA: A new approach to appearance based face representation and recognition [J], IEEE Transaction Pattern Analysis and Machine Intelligence. 2004, 26 (1): 131-137.
[10]S. Chen, Y. Zhu, Sub-pattern based principal component analysis, Pattern Recognition 37 (2004) 1081–1083.
[11] Rajkiran G, Vijayan K.A, An improved face recognition technique based on modular PCA approach, Pattern Recognition Letters 25 (2004) 429–436
[12] J Shawe Taylor, N C fistianini. Kernel methods for pattern analysis, In: Cambridge University Press, Cambridge, 2004.
[13]Yang M H. Kernel Eigenfaces Vs Kernel Fisherfaces: Face recognition using kernel methods [A]. In: Proceedings of Fifth IEEE International Conference on Automatic Face and Gesture Recognition [C], Washington, DC, USA, 2002: 215-220.
[14]B SCHOLKOPF,A SMOLA and K R MULLER.Nonlinear component analysis as a kernel eigenvalue problem [J].Neural Computer, 1998, 10(2):1299-13 19.
[15]贺云辉,赵力,邹采荣.基于KPCA及最佳鉴别独立分量的人脸识别方法[J].应用科学学报, 2005, 23(6): 551-556.
[16]李君宝,潘正祥.一种基于核的监督流形学习算法[J].模式识别与人工智能2008.21(3).
[17] M Belkin and P Niyogi. Laplacian Eigenmaps and Spectral Techniques for Embedding and Clustering. In: Advances in Neural Information Processing System 15.Vancouver, British Columbia,Canada,2001.
[18] Sam T Roweis and Lawrence K. Saul. Nonlinear Dimensionality Reduction by Locally Linear Embedding. In: Science, 2000: 2323-2326.
[19]X. He, S. Yan, Y. Hu, P. Niyogi, and H.-J. Zhang,“Face recognition using laplacianfaces,”IEEE Trans. Pattern Anal. Mach. Intell., vol. 27, no. 3, pp. 328–340, Mar. 2005.
[20]朱明旱,罗大庸. 2DFLD与LPP相结合的人脸和表情识别方法[J].模式识别与人工2009.22(1) .
[21]汪曦,鲁继文,薛延学. LPP算法和DLPP算法在掌纹识别中的应用研究[J].计算机工程与科学2008.30(5) .
[22]马千驰,余国先,钟鸿鹏.一种增强的局部保持投影方法[J].计算机工程与应用2010.46(10).
[23]申中华,潘永惠,王士同.有监督的局部保留投影降维算法[J].模式识别与人工智能2008.21(2).
[24] Feng Guiyu, Hu Dewen, Zhou Zongtan.. A Direct Locality Preserving Projections (DLPP) Algorithm for Image Recognition [J]. IEEE Neural Process Letters. 2008, 27 (3):247-255.
[25]祝磊,朱善安.基于2维保局投影的人脸识别[J].中国图象图形学报2007.12(11) .
[26]申中华,潘永惠,王士同.基于2维保局投影的人脸识别[J].模式识别与人工智能2008.21(2) .
[27]周激流,张哗.人脸识别理论研究进展[J]计算机辅助设计与图像学报.1999,ll(2):23.29.
[28]Erik.Hjellnas, Boon. KeeLow. Face Detcetion: A Survey [J]. Computer Vision and Image Understanding. 2001, 83:236-274.
[29]Y.Amit, D.Geman,B.Jedynak. Efficient focusing and face detection [J]. Face Recognition. 1998.
[30]梁路宏,艾海舟,徐光佑.人脸检测研究综述[J].计算机学报. 2002.
[31]D.Demers, GW. Cottrell. Nonlinear Dimensionality Reduetion [A]. Advances in Neural Information Processing System[C], San Mateo, CA: Morgan Kaufinann. 1993, 580-587.
[32] K. Fukunaga. Introduction to Statistical Pattern Recognition [M], 2nd Edition, New York; Academic Press, 1990.
[33] S. SWilks. Mathematical Statistics [M], New York: Wiley, 1963.
[34] J.M. Lattin, J. Douglas Carroll,Paul E Green. Analyzing Multivariate Data[M],北京:机械出版社, 2003.
[35] M. Turk, A. Pent land. Face Recognition using Eigenfaces [J]. In Proceeding of the IEEE Computer Society Conference on Computer Vision and Pattern Recognition, 1991: 586-591.
[36] A. Pentland, B. Moghaddam, T. Stamer. View-based Modular Eigenspaces for Face Recognition [J]. In Proceeding of the IEEE Computer Society Conference on Computer Vision and Pattern Recognition, 1994:84-91.
[37] Yang-J, Yang-J-Y David Zhang,Jin—Z. KPCA plus LDA: A Complete Kernel Fisher Discriminant Framework for Feature Extraction and Recognition, PAMI (27), No.2, February 2005, PP. 230-244.
[38] J. Yang, D. Zhang, A. F. Frangi, J. Y. Yang. Two-Dimensional PCA: A New Approach to Appearance-Based Face Representation and Recognition [J]. IEEE Transaction on Pattern Analysis And Machine Intelligence, 2004, 26(1): 131-137.
[39] J.Yang, J. Y. Yang. From Image Vector to Martrix: A Straightforward Image Projection Technique-IMPCA vs. PCA [J]. Pattern Recognition, 2002, 35(9):1997-1999.
[40] Sam T Roweis and Lawrence K Saul. Nonlimar Dimensionality Reduction by Locally Linear Embedding. In: Science, 2000: 2323-2326.
[41] M Belkin and P Niyogi. Laplacian Eigenmaps, and Spectral Techniques for Embedding and Clustering. In: Advances in Nemm Information Processing System 15.Vancouver, British Columbia, Canada, 2001.
[42]边肇其,张学工.模式识别.北京:清华大学出版社, 2000.
[43]张志伟,杨帆,夏克文,杨瑞霞.一种有监督的LPP算法及其在人脸识别中的应用[J].电子与信息学报2008.30(3).
[44]祝磊,朱善安.基于2维保局投影的人脸识别[J].中国图象图形学报2007.12(11).
[45]杜海顺李玉玲汪凤泉张帆.一种邻域保持判别嵌入人脸识别方法[J]仪器仪表学报;2008.03.
[46] Yale Univ. Face Database, http://cvc.yale.edu/projects/yalefaces/yalefaces.html.
[47]杜海顺柴秀丽汪凤泉张帆.一种基于双向2DLDA特征融合的人脸识别方法[J]仪器仪表学报;2009.09.
[2] Shakhnarovich G, Moghaddam B. Face Recognition in Subspaces [M]. New York: Sp ringer-Verlag, 2004.
[3] TURK M,PENTLAND A P. Face recognition using Eigen faces [C].IEEE Conference on Computer Vision and Pattern Recognition. Los Akanitos, 1991:586-591.
[4] ZHUANG Z M, ZHANG A N, LI F L. Based on an optimized LDA algorithm for face recognition [J]. Journal of Electronics & Information Technology, 2007, 29(9):2047-2049.
[5] Zhou DL, Zhao DB. Face recognition based on singular value decomposition and discriminant KL projection.In: Journal of Software, 2003, 14(4):783-789.
[6] AH, and E Oja. Independent component analysis: algorithms and applications. In: Neural Networks, 2000, 13(4):411-430.
[7] Pcnio S Pencv. Local feature analysis:a statistical theory for information representation and transmission.New York:Rockefeller University,1998.
[8] Borg I and Grocnen P Modern.Multidimensional Scaling.In:Theory and Applications, New York:Springer-Verlag, 1997.
[9] YANG J, ZHANG D, and FRANGI A.F, et al. Two-dimensional PCA: A new approach to appearance based face representation and recognition [J], IEEE Transaction Pattern Analysis and Machine Intelligence. 2004, 26 (1): 131-137.
[10]S. Chen, Y. Zhu, Sub-pattern based principal component analysis, Pattern Recognition 37 (2004) 1081–1083.
[11] Rajkiran G, Vijayan K.A, An improved face recognition technique based on modular PCA approach, Pattern Recognition Letters 25 (2004) 429–436
[12] J Shawe Taylor, N C fistianini. Kernel methods for pattern analysis, In: Cambridge University Press, Cambridge, 2004.
[13]Yang M H. Kernel Eigenfaces Vs Kernel Fisherfaces: Face recognition using kernel methods [A]. In: Proceedings of Fifth IEEE International Conference on Automatic Face and Gesture Recognition [C], Washington, DC, USA, 2002: 215-220.
[14]B SCHOLKOPF,A SMOLA and K R MULLER.Nonlinear component analysis as a kernel eigenvalue problem [J].Neural Computer, 1998, 10(2):1299-13 19.
[15]贺云辉,赵力,邹采荣.基于KPCA及最佳鉴别独立分量的人脸识别方法[J].应用科学学报, 2005, 23(6): 551-556.
[16]李君宝,潘正祥.一种基于核的监督流形学习算法[J].模式识别与人工智能2008.21(3).
[17] M Belkin and P Niyogi. Laplacian Eigenmaps and Spectral Techniques for Embedding and Clustering. In: Advances in Neural Information Processing System 15.Vancouver, British Columbia,Canada,2001.
[18] Sam T Roweis and Lawrence K. Saul. Nonlinear Dimensionality Reduction by Locally Linear Embedding. In: Science, 2000: 2323-2326.
[19]X. He, S. Yan, Y. Hu, P. Niyogi, and H.-J. Zhang,“Face recognition using laplacianfaces,”IEEE Trans. Pattern Anal. Mach. Intell., vol. 27, no. 3, pp. 328–340, Mar. 2005.
[20]朱明旱,罗大庸. 2DFLD与LPP相结合的人脸和表情识别方法[J].模式识别与人工2009.22(1) .
[21]汪曦,鲁继文,薛延学. LPP算法和DLPP算法在掌纹识别中的应用研究[J].计算机工程与科学2008.30(5) .
[22]马千驰,余国先,钟鸿鹏.一种增强的局部保持投影方法[J].计算机工程与应用2010.46(10).
[23]申中华,潘永惠,王士同.有监督的局部保留投影降维算法[J].模式识别与人工智能2008.21(2).
[24] Feng Guiyu, Hu Dewen, Zhou Zongtan.. A Direct Locality Preserving Projections (DLPP) Algorithm for Image Recognition [J]. IEEE Neural Process Letters. 2008, 27 (3):247-255.
[25]祝磊,朱善安.基于2维保局投影的人脸识别[J].中国图象图形学报2007.12(11) .
[26]申中华,潘永惠,王士同.基于2维保局投影的人脸识别[J].模式识别与人工智能2008.21(2) .
[27]周激流,张哗.人脸识别理论研究进展[J]计算机辅助设计与图像学报.1999,ll(2):23.29.
[28]Erik.Hjellnas, Boon. KeeLow. Face Detcetion: A Survey [J]. Computer Vision and Image Understanding. 2001, 83:236-274.
[29]Y.Amit, D.Geman,B.Jedynak. Efficient focusing and face detection [J]. Face Recognition. 1998.
[30]梁路宏,艾海舟,徐光佑.人脸检测研究综述[J].计算机学报. 2002.
[31]D.Demers, GW. Cottrell. Nonlinear Dimensionality Reduetion [A]. Advances in Neural Information Processing System[C], San Mateo, CA: Morgan Kaufinann. 1993, 580-587.
[32] K. Fukunaga. Introduction to Statistical Pattern Recognition [M], 2nd Edition, New York; Academic Press, 1990.
[33] S. SWilks. Mathematical Statistics [M], New York: Wiley, 1963.
[34] J.M. Lattin, J. Douglas Carroll,Paul E Green. Analyzing Multivariate Data[M],北京:机械出版社, 2003.
[35] M. Turk, A. Pent land. Face Recognition using Eigenfaces [J]. In Proceeding of the IEEE Computer Society Conference on Computer Vision and Pattern Recognition, 1991: 586-591.
[36] A. Pentland, B. Moghaddam, T. Stamer. View-based Modular Eigenspaces for Face Recognition [J]. In Proceeding of the IEEE Computer Society Conference on Computer Vision and Pattern Recognition, 1994:84-91.
[37] Yang-J, Yang-J-Y David Zhang,Jin—Z. KPCA plus LDA: A Complete Kernel Fisher Discriminant Framework for Feature Extraction and Recognition, PAMI (27), No.2, February 2005, PP. 230-244.
[38] J. Yang, D. Zhang, A. F. Frangi, J. Y. Yang. Two-Dimensional PCA: A New Approach to Appearance-Based Face Representation and Recognition [J]. IEEE Transaction on Pattern Analysis And Machine Intelligence, 2004, 26(1): 131-137.
[39] J.Yang, J. Y. Yang. From Image Vector to Martrix: A Straightforward Image Projection Technique-IMPCA vs. PCA [J]. Pattern Recognition, 2002, 35(9):1997-1999.
[40] Sam T Roweis and Lawrence K Saul. Nonlimar Dimensionality Reduction by Locally Linear Embedding. In: Science, 2000: 2323-2326.
[41] M Belkin and P Niyogi. Laplacian Eigenmaps, and Spectral Techniques for Embedding and Clustering. In: Advances in Nemm Information Processing System 15.Vancouver, British Columbia, Canada, 2001.
[42]边肇其,张学工.模式识别.北京:清华大学出版社, 2000.
[43]张志伟,杨帆,夏克文,杨瑞霞.一种有监督的LPP算法及其在人脸识别中的应用[J].电子与信息学报2008.30(3).
[44]祝磊,朱善安.基于2维保局投影的人脸识别[J].中国图象图形学报2007.12(11).
[45]杜海顺李玉玲汪凤泉张帆.一种邻域保持判别嵌入人脸识别方法[J]仪器仪表学报;2008.03.
[46] Yale Univ. Face Database, http://cvc.yale.edu/projects/yalefaces/yalefaces.html.
[47]杜海顺柴秀丽汪凤泉张帆.一种基于双向2DLDA特征融合的人脸识别方法[J]仪器仪表学报;2009.09.