图像处理在人脸识别中的应用
|
摘要
作为生物特征识别中最自然最直接的手段,人脸识别技术受到越来越多的关注。基于可见光图像的人脸识别系统目前已取得一些成果,但是其不仅不适合于对伪装脸的识别,而且性能还受到光照变化的影响,尤其当照明非均匀、光照昏暗或者在户外时,识别率会明显降低。基于红外的人脸图像识别系统虽然对光照的鲁棒性较好,但是当待识别对象戴有眼镜时,系统识别性能会骤然下降。基于上述问题,本文将如何利用图像融合技术来提高人脸识别系统的整体识别性能作为研究对象。将基于奇异值分解(SVD)图像分层的技术应用于多模式图像融合,然后基于主成分分析(PCA)方法利用融合图像进行人脸识别。主要内容如下:
(1)研究实现人脸图像归一化的方法。首先实现图像的灰度归一化,然后进行人眼定位,利用人眼位置信息实现人脸图像的几何归一化,得到标准人脸。
(2)研究人脸图像SVD分解和基于PCA的人脸识别算法。介绍了SVD分解的原理以及SVD分解在人脸识别中的应用,并详细阐述了基于奇异值分解的人脸特征提取的方法以及基于PCA实现人脸识别的具体方法和步骤。
(3)研究基于红外与可见光图像融合的人脸识别方法,主要包括两个部分:一是基于像素级融合的人脸识别研究;二是基于决策级融合的人脸识别效果研究。前者主要讨论基于小波分解和基于奇异值分解(SVD)的红外人脸图像和可见光人脸图像的融合技术,然后将采用两种融合处理得到的结果图像应用于人脸识别;后者实现可见光图像和红外图像两种识别结果的决策级融合。论文对比分析了几种融合算法的实验结果,并验证了该算法的有效性。
Most recently, as the most natural and explicit approach for biological feature recognition, face recognition has attracted increasing attentions. Although face recognition approaches based on the visual light spectrum have gained some success,it not only unsuitable for the disguised face, but also the performance may be affected by the change of the lighting, especially when the lighting is uneven or dim, or in the outdoor circumstances, the recognition rate may decline obviously. Face recognition system based on infrared (IR) have better performance in terms of robustness.
However, when the people put on the glasses, the performance of system recognition shows a shape decrease. Based on all the issues mentioned above, in the thesis, the study focus on multi spectral image fusion technology that is utilized to improve the overall performance of the Face recognition system. the image slicing technology based on singular value decomposition (SVD) is used to the fusion of multi-pattern image, then the fusion image is used to recognize the faces based on the principal component analysis (PCA) method.
(1)Approach of normalization for face image the is studied. Firstly, gray normalization of the face image is realized, then the eye location process is carried out, the standard face is obtained by using geometric normalization the based on the position information of the eyes.
(2)Further, the singular value decomposition (SVD) of and the face recognition algorithm based on PCA are studied. Principles and applications of the SVD for the face image are introduced, the method of obtaining features of face and steps for face recognition based on PCA are illustrated as well.
(3)Image slicing approaches is utilized to realize the fusion of visible and light infrared based on the energy of the image. Firstly, the SVD decomposition of layers according to different energy , i.e., low, high and ultrahigh resolution layers ,respectively. Besides, in each layer, the fusion strategies is used according to the different performance features. Last, the face recognition algorithm is realized through the fusion of the image. In the thesis, several experimental results of fusion algorithm are compared, the effectiveness of the algorithm is verified as well.
(1)研究实现人脸图像归一化的方法。首先实现图像的灰度归一化,然后进行人眼定位,利用人眼位置信息实现人脸图像的几何归一化,得到标准人脸。
(2)研究人脸图像SVD分解和基于PCA的人脸识别算法。介绍了SVD分解的原理以及SVD分解在人脸识别中的应用,并详细阐述了基于奇异值分解的人脸特征提取的方法以及基于PCA实现人脸识别的具体方法和步骤。
(3)研究基于红外与可见光图像融合的人脸识别方法,主要包括两个部分:一是基于像素级融合的人脸识别研究;二是基于决策级融合的人脸识别效果研究。前者主要讨论基于小波分解和基于奇异值分解(SVD)的红外人脸图像和可见光人脸图像的融合技术,然后将采用两种融合处理得到的结果图像应用于人脸识别;后者实现可见光图像和红外图像两种识别结果的决策级融合。论文对比分析了几种融合算法的实验结果,并验证了该算法的有效性。
Most recently, as the most natural and explicit approach for biological feature recognition, face recognition has attracted increasing attentions. Although face recognition approaches based on the visual light spectrum have gained some success,it not only unsuitable for the disguised face, but also the performance may be affected by the change of the lighting, especially when the lighting is uneven or dim, or in the outdoor circumstances, the recognition rate may decline obviously. Face recognition system based on infrared (IR) have better performance in terms of robustness.
However, when the people put on the glasses, the performance of system recognition shows a shape decrease. Based on all the issues mentioned above, in the thesis, the study focus on multi spectral image fusion technology that is utilized to improve the overall performance of the Face recognition system. the image slicing technology based on singular value decomposition (SVD) is used to the fusion of multi-pattern image, then the fusion image is used to recognize the faces based on the principal component analysis (PCA) method.
(1)Approach of normalization for face image the is studied. Firstly, gray normalization of the face image is realized, then the eye location process is carried out, the standard face is obtained by using geometric normalization the based on the position information of the eyes.
(2)Further, the singular value decomposition (SVD) of and the face recognition algorithm based on PCA are studied. Principles and applications of the SVD for the face image are introduced, the method of obtaining features of face and steps for face recognition based on PCA are illustrated as well.
(3)Image slicing approaches is utilized to realize the fusion of visible and light infrared based on the energy of the image. Firstly, the SVD decomposition of layers according to different energy , i.e., low, high and ultrahigh resolution layers ,respectively. Besides, in each layer, the fusion strategies is used according to the different performance features. Last, the face recognition algorithm is realized through the fusion of the image. In the thesis, several experimental results of fusion algorithm are compared, the effectiveness of the algorithm is verified as well.
引文
[1]王伟,马建光.人脸识别常用方法及其发展现状.自动检测技术,2002,21(l): 55-88.
[2]张翠平,苏光大.人脸识别技术综述.中国图像图形学报,2000,S (ll):885-894.
[3]周激流,张哗.人脸识别理论研究进展.计算机辅助设计与图形学学报,1993,11(2):51-61.
[4] Peng H,Zhang D. Dualeigen space method for human face recognition[J]. Electronics Letters,1997,33(4).283-284.
[5] Sirovich L, Kirby M. Low-dimensional procedure for the characterization of human faces. Journal of the Optical Society of America A. Mar. 1987. 4(3). 519-524.
[6] Turk M A, Pentland A P. Recognition in face space. Intelligent Robots and Computer Vision IX: Algorithms and Techniques. SPIE. Nov. 1990. 43-54.
[7] Yu H, Yang J. A direct lda algorithm for high-dimensional data with application to face recognition. Pattern Recognition. 2001. 34(10). 2067-2070.
[8] Chen L F, Liao H, Ko M T, et al. A new lda-based face recognition system which can solve the small sample size problem. Pattern recognition. Oct. 2000. 33(10). 1713-1726.
[9]刘小军,王东峰,张丽飞,时永刚,邹谋炎,一种基于奇异值分解和隐马尔可夫模型的人脸识别方法,计算机学报,2003.26(3),85-59
[10] A.V Nefian,M. H.,Hayes,Hidden Markov models for face recognition,Proc. IEEE Conf. Acoustic Speech,Signal Proeess.1988,5, 2721-2724.
[11] A.V Nefian,M.H.,Hayes,A nembedded HMM-based approach for face detection and recognition,Proc.IEEE Int Conf. Acoustic Speech,Signal Proeess.1999,6,3553-3556
[12] Steve Lwarence,et al. Face recognition: a convolution neural-network approach.IEEE Trnasactions on Neural networks,1997,8(1):98-113
[13] Flming M,Cottrell G Categorization of a feesusingun super VI Sedefature extraction. Proceedings of the International Conefrenee on Neural Networks. Caliofnria Univ. AanDiego,CA,USA,1990,65-70
[14] Intrator N,Reisefld D,Yeshurun Y. Face recognition supervised/unsupervised hybrid network.Pattern Reeognition Letters.1996,17(1):67-76
[15] Lin S H,Hung S Y. Lin L J.Face recognition/detection by probability Decision-based neural network.IEEE Transactions on Neural Networks,1997,8(0):114-132
[16]甘俊英,张有为.一种基于奇异值特征的神经络人脸识别新途径.电子学报,2004,32(1):170-173
[17] Jiang X, Mandal B, Kot A. Face recognition based on discriminant evaluation in the whole space. ICASSP 2007, IEEE. Apr. 2007. 245-248.
[18] Lai J h,Yuen p C,Feng G C. Face recognition using holistie Fourier invariant features.Pattern Recognition,2001,34(l):95-109.
[19] Liu J,Weehsler H. Evolutiona pursuit and it salication to face recognition[J].IEEE Transactions on Pattern Analysis and machine Intelligence,2000,22(6):570-582.
[20] Rama Chellappa,et. al. Human and machine recognition of faces: a survey. Proc. IEEE,1995,83(5):705-740.
[21]王骅.彩色图像中人脸检测与跟踪研究.南京理工大学硕士学位论文.2009,6.
[22] Rama Chellappa,et. al. Human and machine recognition of faces: a survey. Proc. IEEE,1995,83(5):705-740.
[23]梁毅雄,龚卫国,潘英俊,李伟红,刘嘉敏,张红梅.基于奇异值分解的人脸识别方法[J].光学精密工程,2004,12(5):543-549.
[24] Turk M and Pentland A. Face processing: a Models for recognition. Proc. Intelligent Robots and Computer vision VIII, SPIE,1989,1, 192:22-32.
[25] Hong Z Q. Algebraic feature extraction of image for recognition. Pattern Recognition, 1991,24(3):211-219.
[26] Cheng Yong-qiang, Liu Ke,Yang Jin-yu. Arobust algebraic method for human face recognition[A]. In:Internation Conference on Pattern Recognition. Hague, Netherlands,1992:221-224.
[27] Tian Yuan,Tan Tie-niu, Wang Yun-hong. Do singular values contain adequate information for face recognition. Pattern recognition,2003,36(6): 649-655.
[28]高全学,梁彦,潘泉等. SVD用于人脸识别存在的问题及解决方法.中国图像图形学报.2006,11(12):1784-1791.
[29]李竹林,赵红漫,赵宗涛,马燕.改进的奇异值求解算法及其在目标识别中的应用.计算机工程,2005,31(21):151-152·
[30]李竹林,赵宗涛.奇异值特征在目标识别中的应用.微电子学与计算机,2004,30(11):27-29.
[31]陈献忠,苏庆刚,王耀明.应用于人脸识别的结合SVD变换的图像类特征提取算法.计算机应用与软件,2010,27(9):231-233.
[32]杜干,朱雯君.基于局部奇异值分解和模糊决策的人脸识别方法.中国图象图形学报,2006,11(10):2456-1459.
[33]甘俊英,何国辉,梁宇.基于局部奇异值对称平均的人脸识别方法.计算机工程,2005,31(17):146-148.
[34] K.Etemad and R.Chellappa. Discriminant Analysis for Recognition of Human Face Images. Journal of Optical Society of America,Vol.14,No.8.1997.1724-1733.
[35]P.Comon,“Independent component analysis,a new concept”Signal Processing Vol.36,No.3.1993.287-314。.
[36] M.S. Bartlett, J. R. Movellan, T.J. Sejnowski. Face recognition by independent component analysis. IEEE Trans. Neural Networks, 2002,13(6): 1450-1464.
[37]夏明革,何友,唐小明等.多传感器图像融合综述.电光与控制,2002,9(4):17.
[38]张大明.像素级多分辨率遥感图像融合算法的研究.安徽大学硕士学位论文,2005.
[39]张彬,郑永果,东野长磊.基于小波帧变换的图像融合方法.武汉理工大学学报,2010,32(20):156-159
[40]黄大庆,李慧娟.基于小波变换和自适应加权的无人机侦察图像融合.遥测遥控,2010,31(5):54-60
[41]王海晖,彭嘉雄.基于多小波变换的图像融合研究.中国图象图形学报,2004,9(8):1002-1007.
[42] Chao R,Zhang K,Li Y J. An image fusion algorithm using wavelet transform.Aiea Electronical Siniea,2004,32.750-753.
[43] Lallier E,Farooq M A. Real time Pixel-level based image fusion via adaptive weight averaging.ISIF2000,3-13.
[44] Burt P J,Kolczynski R J. Enhanced image capture through fusion. The 4th Intl. Conf.On Conputer Vision,1993.173-182.
[45] Gyaourova A,Bebisq Pavlidisl. Fusion of infrared and visible images for face recognition. Pajdla T and Matas J(Eds):ECCV2004,456-468.
[46]周爱平,梁久祯.基于二代Curvelet变换与MPCA的可见光与红外图像融合.计算机应用,2010,30(11):3011-3014.
[47]甘俊英,何国辉,梁宇.基于局部奇异值对称平均的人脸识别方法.计算机工程,2005,31(17):146-148.
[48]叶剑华,刘正光.多模态人脸识别融合方法比较研究.计算机工程与应用. 2009,45(19):153-156
[2]张翠平,苏光大.人脸识别技术综述.中国图像图形学报,2000,S (ll):885-894.
[3]周激流,张哗.人脸识别理论研究进展.计算机辅助设计与图形学学报,1993,11(2):51-61.
[4] Peng H,Zhang D. Dualeigen space method for human face recognition[J]. Electronics Letters,1997,33(4).283-284.
[5] Sirovich L, Kirby M. Low-dimensional procedure for the characterization of human faces. Journal of the Optical Society of America A. Mar. 1987. 4(3). 519-524.
[6] Turk M A, Pentland A P. Recognition in face space. Intelligent Robots and Computer Vision IX: Algorithms and Techniques. SPIE. Nov. 1990. 43-54.
[7] Yu H, Yang J. A direct lda algorithm for high-dimensional data with application to face recognition. Pattern Recognition. 2001. 34(10). 2067-2070.
[8] Chen L F, Liao H, Ko M T, et al. A new lda-based face recognition system which can solve the small sample size problem. Pattern recognition. Oct. 2000. 33(10). 1713-1726.
[9]刘小军,王东峰,张丽飞,时永刚,邹谋炎,一种基于奇异值分解和隐马尔可夫模型的人脸识别方法,计算机学报,2003.26(3),85-59
[10] A.V Nefian,M. H.,Hayes,Hidden Markov models for face recognition,Proc. IEEE Conf. Acoustic Speech,Signal Proeess.1988,5, 2721-2724.
[11] A.V Nefian,M.H.,Hayes,A nembedded HMM-based approach for face detection and recognition,Proc.IEEE Int Conf. Acoustic Speech,Signal Proeess.1999,6,3553-3556
[12] Steve Lwarence,et al. Face recognition: a convolution neural-network approach.IEEE Trnasactions on Neural networks,1997,8(1):98-113
[13] Flming M,Cottrell G Categorization of a feesusingun super VI Sedefature extraction. Proceedings of the International Conefrenee on Neural Networks. Caliofnria Univ. AanDiego,CA,USA,1990,65-70
[14] Intrator N,Reisefld D,Yeshurun Y. Face recognition supervised/unsupervised hybrid network.Pattern Reeognition Letters.1996,17(1):67-76
[15] Lin S H,Hung S Y. Lin L J.Face recognition/detection by probability Decision-based neural network.IEEE Transactions on Neural Networks,1997,8(0):114-132
[16]甘俊英,张有为.一种基于奇异值特征的神经络人脸识别新途径.电子学报,2004,32(1):170-173
[17] Jiang X, Mandal B, Kot A. Face recognition based on discriminant evaluation in the whole space. ICASSP 2007, IEEE. Apr. 2007. 245-248.
[18] Lai J h,Yuen p C,Feng G C. Face recognition using holistie Fourier invariant features.Pattern Recognition,2001,34(l):95-109.
[19] Liu J,Weehsler H. Evolutiona pursuit and it salication to face recognition[J].IEEE Transactions on Pattern Analysis and machine Intelligence,2000,22(6):570-582.
[20] Rama Chellappa,et. al. Human and machine recognition of faces: a survey. Proc. IEEE,1995,83(5):705-740.
[21]王骅.彩色图像中人脸检测与跟踪研究.南京理工大学硕士学位论文.2009,6.
[22] Rama Chellappa,et. al. Human and machine recognition of faces: a survey. Proc. IEEE,1995,83(5):705-740.
[23]梁毅雄,龚卫国,潘英俊,李伟红,刘嘉敏,张红梅.基于奇异值分解的人脸识别方法[J].光学精密工程,2004,12(5):543-549.
[24] Turk M and Pentland A. Face processing: a Models for recognition. Proc. Intelligent Robots and Computer vision VIII, SPIE,1989,1, 192:22-32.
[25] Hong Z Q. Algebraic feature extraction of image for recognition. Pattern Recognition, 1991,24(3):211-219.
[26] Cheng Yong-qiang, Liu Ke,Yang Jin-yu. Arobust algebraic method for human face recognition[A]. In:Internation Conference on Pattern Recognition. Hague, Netherlands,1992:221-224.
[27] Tian Yuan,Tan Tie-niu, Wang Yun-hong. Do singular values contain adequate information for face recognition. Pattern recognition,2003,36(6): 649-655.
[28]高全学,梁彦,潘泉等. SVD用于人脸识别存在的问题及解决方法.中国图像图形学报.2006,11(12):1784-1791.
[29]李竹林,赵红漫,赵宗涛,马燕.改进的奇异值求解算法及其在目标识别中的应用.计算机工程,2005,31(21):151-152·
[30]李竹林,赵宗涛.奇异值特征在目标识别中的应用.微电子学与计算机,2004,30(11):27-29.
[31]陈献忠,苏庆刚,王耀明.应用于人脸识别的结合SVD变换的图像类特征提取算法.计算机应用与软件,2010,27(9):231-233.
[32]杜干,朱雯君.基于局部奇异值分解和模糊决策的人脸识别方法.中国图象图形学报,2006,11(10):2456-1459.
[33]甘俊英,何国辉,梁宇.基于局部奇异值对称平均的人脸识别方法.计算机工程,2005,31(17):146-148.
[34] K.Etemad and R.Chellappa. Discriminant Analysis for Recognition of Human Face Images. Journal of Optical Society of America,Vol.14,No.8.1997.1724-1733.
[35]P.Comon,“Independent component analysis,a new concept”Signal Processing Vol.36,No.3.1993.287-314。.
[36] M.S. Bartlett, J. R. Movellan, T.J. Sejnowski. Face recognition by independent component analysis. IEEE Trans. Neural Networks, 2002,13(6): 1450-1464.
[37]夏明革,何友,唐小明等.多传感器图像融合综述.电光与控制,2002,9(4):17.
[38]张大明.像素级多分辨率遥感图像融合算法的研究.安徽大学硕士学位论文,2005.
[39]张彬,郑永果,东野长磊.基于小波帧变换的图像融合方法.武汉理工大学学报,2010,32(20):156-159
[40]黄大庆,李慧娟.基于小波变换和自适应加权的无人机侦察图像融合.遥测遥控,2010,31(5):54-60
[41]王海晖,彭嘉雄.基于多小波变换的图像融合研究.中国图象图形学报,2004,9(8):1002-1007.
[42] Chao R,Zhang K,Li Y J. An image fusion algorithm using wavelet transform.Aiea Electronical Siniea,2004,32.750-753.
[43] Lallier E,Farooq M A. Real time Pixel-level based image fusion via adaptive weight averaging.ISIF2000,3-13.
[44] Burt P J,Kolczynski R J. Enhanced image capture through fusion. The 4th Intl. Conf.On Conputer Vision,1993.173-182.
[45] Gyaourova A,Bebisq Pavlidisl. Fusion of infrared and visible images for face recognition. Pajdla T and Matas J(Eds):ECCV2004,456-468.
[46]周爱平,梁久祯.基于二代Curvelet变换与MPCA的可见光与红外图像融合.计算机应用,2010,30(11):3011-3014.
[47]甘俊英,何国辉,梁宇.基于局部奇异值对称平均的人脸识别方法.计算机工程,2005,31(17):146-148.
[48]叶剑华,刘正光.多模态人脸识别融合方法比较研究.计算机工程与应用. 2009,45(19):153-156