人脸识别中基于TV模型的光照不变量提取
|
摘要
人脸识别技术在实际中已经得到广泛的应用,而现在大多数人脸识别算法都对光照比较敏感,光照问题已经成为影响识别结果最主要的因素之一。处理光照问题最常用的一种方法是寻找具有光照不变特性的不变量来描述光照条件下的图像,这些光照不变量主要包括图像的高频和边缘信息。而提取图像高频和边缘信息的一种有效办法为TV模型。
基于TV模型的光照不变量提取方法能够在保持图像边缘的基础上,充分地提取用于识别的人脸高频细节特征,但也存在对光照不变量划分不够精确以及参数优化过于随机的问题,针对以上问题引出了基于G范数的TV模型,并在此基础上提出了一种基于自适应参数的G范数TV模型,此模型可以对人脸细节特征进行更精确的划分,得到更有利于识别的光照不变量。针对TV模型存在全局化、常值区域等问题,提出了基于TV模型和Contourlet变换相结合的方法,算法充分利用了Contourlet变换局部性、多方向性和TV模型保持边缘的优点,能有效地提取用来识别的人脸光照不变量。
在Yale-B人脸数据库上,本文提出的两种模型的平均识别率相对于直接使用PCA+LDA分别提高了40.11%和40.65%,在最恶劣光照条件下都提高了86.87%。相对于传统的TV模型,平均识别率也分别提高了1.91%和2.45%,在最恶劣光照条件下分别提高了1.41%和1.95%,并且基于自适应参数的G范数TV模型还有效地减少了参数优化的时间。这表明本文提出的算法能够较好地改进传统TV模型的缺点,是非常有效的光照不变量提取方法。
Face Recognition Technology has been widely used in practice, but most face recognition algorithms are very sensitive to light, illumination has became one of the most decisive factors for the recognition result. One of most common method for illumination problems is looking for a illumination normalization, the illumination normalization mainly include high-frequency and edge information of image, it is insensitive to light and can describe image under different light conditions. TV model is an effective method for extracting high-frequency and edge information of image.
TV model can well maintain the edge of image and extract most useful high-frequency details for face recognition, but the illumination normalization obtained by TV model based L norm is not precise, and the parameters of this TV model is very random. To solve the above problems, we lead to the TV model based on G-norm, and propose an adaptive G-norm TV model, this model can obtain more accurate illumination normalization, and it is more conductive to recognize. To solve the globalization and the constant area of TV model, we propose a new model combing TV model and Contourlet transform, This algorithm takes full advantage of localization and the multi-dimensional of Contourlet transform and the edge-preserve ability of total variation models, it can effectively obtain the face illumination normalization for the face recognition.
Experiments are carried out Upon the Yale-B database demonstrate that the proposed method achieves satisfactory recognition rates under varying illumination conditions. Compared with directly method of PCA+LDA, the proposed method has an average recognition ratio increase 40.11% and 40.65%. and all increase 86.87% in the worst light conditions; Compared with the traditional TV model, has an average recognition ratio increase 1.91% and 2.45%, and respectively increase 1.41% and 1.95%, so the proposed model are effective method for face illumination normalization.
基于TV模型的光照不变量提取方法能够在保持图像边缘的基础上,充分地提取用于识别的人脸高频细节特征,但也存在对光照不变量划分不够精确以及参数优化过于随机的问题,针对以上问题引出了基于G范数的TV模型,并在此基础上提出了一种基于自适应参数的G范数TV模型,此模型可以对人脸细节特征进行更精确的划分,得到更有利于识别的光照不变量。针对TV模型存在全局化、常值区域等问题,提出了基于TV模型和Contourlet变换相结合的方法,算法充分利用了Contourlet变换局部性、多方向性和TV模型保持边缘的优点,能有效地提取用来识别的人脸光照不变量。
在Yale-B人脸数据库上,本文提出的两种模型的平均识别率相对于直接使用PCA+LDA分别提高了40.11%和40.65%,在最恶劣光照条件下都提高了86.87%。相对于传统的TV模型,平均识别率也分别提高了1.91%和2.45%,在最恶劣光照条件下分别提高了1.41%和1.95%,并且基于自适应参数的G范数TV模型还有效地减少了参数优化的时间。这表明本文提出的算法能够较好地改进传统TV模型的缺点,是非常有效的光照不变量提取方法。
Face Recognition Technology has been widely used in practice, but most face recognition algorithms are very sensitive to light, illumination has became one of the most decisive factors for the recognition result. One of most common method for illumination problems is looking for a illumination normalization, the illumination normalization mainly include high-frequency and edge information of image, it is insensitive to light and can describe image under different light conditions. TV model is an effective method for extracting high-frequency and edge information of image.
TV model can well maintain the edge of image and extract most useful high-frequency details for face recognition, but the illumination normalization obtained by TV model based L norm is not precise, and the parameters of this TV model is very random. To solve the above problems, we lead to the TV model based on G-norm, and propose an adaptive G-norm TV model, this model can obtain more accurate illumination normalization, and it is more conductive to recognize. To solve the globalization and the constant area of TV model, we propose a new model combing TV model and Contourlet transform, This algorithm takes full advantage of localization and the multi-dimensional of Contourlet transform and the edge-preserve ability of total variation models, it can effectively obtain the face illumination normalization for the face recognition.
Experiments are carried out Upon the Yale-B database demonstrate that the proposed method achieves satisfactory recognition rates under varying illumination conditions. Compared with directly method of PCA+LDA, the proposed method has an average recognition ratio increase 40.11% and 40.65%. and all increase 86.87% in the worst light conditions; Compared with the traditional TV model, has an average recognition ratio increase 1.91% and 2.45%, and respectively increase 1.41% and 1.95%, so the proposed model are effective method for face illumination normalization.
引文
[1] H.Chan, W.W.Bledsoe. A man-machine facial recognition system:some preliminary results.Technical report,Panoramic Research Inc..1965
[2] N. Furl, A. J. O’Toole, and P. J. Phillips,“Face recognition algorithms as models of the other race effect,”Cogn. Sci., vol. 96, pp. 1–19, 2002.
[3] A. Adler and J. Maclean,“Performance comparison of human and automatic face recognition,”in Proc. Biometrics Consortium Conf., Washington, DC, Sep. 20–22, 2004.
[4] C. H. Liu, H. Seetzen, A. M. Burton, and A. Chaudhuri,“Face recognition is robust with incongruent image resolution: Relationship to security video images,”J. Exper. Psychol.,Appl., vol. 9, no. 1, pp. 33–41, Mar. 2003.
[5] R.J. Baron. Mechanisms of Human Facial Recognition. Int.J. Man-Machine Studies,1981, 15, pp:137-178.
[6] L. Wiskott, J.M. Fellous and N. Kruger. Face Recognition by Elastic Bunch Graph Matching. IEEE Trans. Pattern Analysis and Machine Intelligence, 19(7),pp:775-779. 1997.
[7] M. Turk and A. Pentland, "Eigenfaces for Recognition", Journal of cognitive neuroscience, 1991 3(1), pp:71-86.
[8] B.Moghaddam.”Principal manifolds and probabilistic subspaces for visual recognition’,IEEE Trans.PAMI 24(6) 780-788.2002.
[9] B.Moghaddam and A. Pentland.”Probabilistic Visual Learning for Object Detection”.Proc.Int’l Conf.Computer Vision,pp.786-793,1995.
[10] B.Moghaddam and A. Pentland.”Probabilistic Visual Learning for Object Representation”. IEEE Trans.PAMI,vol.20,n0.7,pp.696-710,1997.
[11] Baback Moghaddam,Tony Jebara and Alex Pentland.”Bayesian Face Recognition’.Pattern Recognition Vol.33(2000),pp1771-1782,2000.
[12] P.Belhumeur,J.Hespanha and D.Kriegman.”Eigenfaces vs.Fisherfaces:Recogition using class specific linear projection”. in Processing of Fourth European Conference on Computer Vision ECCV’96,pp45-56,1996.
[13] P.N. Belhumeur, J.Hespanha and D.Kriegman.”Eigenfaces vs.Fisherfaces:Recogition using class specific linear projection”.IEEE Transacitions on Pattern Analysis and Machine Intelligence,Special Issue on Face Recognition,17(7):711-720,1997.
[14] R.Duda,P.Hart,D.stork,”Pattern Classidication”,John Wiley & Sons,NewYork,2001.
[15] H.Yu and H.Yang,”A direct LDA algorithm for high-dimensional data-with application to face recognition’,Pattern Recognition,vol.34,no.10,pp2067-2070,2001.
[16] C.Liu and H.Wechsler,”Enhanced Fisher Linear Discriminant Models for Face Recognition,”Proceeedings of International Conference on Pattern Recognition,vol.2,pp.1368-1372,1998.
[17] L.F.Chen.H.M.Liao,J.C.Lin,M.T.Ko,and G.J.Yu.”Anew LDA-based Face recognition Systerm Which Can Solve the Small Sample Size Problem’.Pattern Recognition,vol.33,no.10,pp1712-1726,2000.
[18] M.H.Yang,N,Ahuja,and D.Kriegman.”Face Recognition Using Kerne Eigenfaces’,Int. Conf on image Processing ,vol.1,pp37-40.,2000.
[19] M.H.Yang.”Kernel Eigenfaces vs Kernel Fisherfaces:Face Recognition Using Kernel Methods’,Proc.Int’l,Conf.Automatic Face Gesture Recognition,pp.215-220,2002.
[20] M.Propp and A.Samal,”Artificial Neural Network Architectures for Face Detection,”Intelligent Eng.Systerms through Artificial Neural Network,vol.2,1992.
[21]A.N.Pajagopalan,K.S.Kumar,J.Karlekar,R.Manivasakan,M.M.Patil,U.B.Desai, P.G.Poonacha,and S.Chaudburi,”Finding Faces in Photographs,”Proceeding of IEEE Conf. Computer Vision,pp640-645,1998.
[22] P.N. Belhumeur and D.J.Kriegman,”what is the set of images of an Object under all Possible Lighting Conditions”IEEE Int’l Conf Computer Vision and Pattern Recognization,2004.
[23] R.Basri and D.Jacobs,”Lambertain Refletance and Linear Subspaces,”Technical Report 2000-172R,NEC Research Inst,2000.
[24] K.K.Sung and T.Poggio,”Example Based Learning for View-Based Human Face Detection,”IEEE Trans.Pattern Analysis and Machine Intelligence,vol.20,no.1,pp39-51,1998.
[25] B.Moghaddam and A.Pentland,”Probabilistic Visual Learning for Object Representation,”IEEE Trans.Pattern Analysis and Machine Intelligence,vol.19.no.17,pp696-710,1997.
[26] R.Brunelli and T.Poggio,”Hyper BF Networks for Real Object Recognition,”Praceedings of International Joint Conf.on Artificial Intelligence,pp1278-1284,1991.
[27] V.Govindaraju,”Locating human faces in photographs,”International Journal of Computer Vision,vol.19,no.2.pp129-146,1996.
[28] A.Shashua,”Geometry and Photometry in 3D Visual Recognition.”Ph.D.MIT,Artificial Intelligence Lab,1992.
[29] D.Jacobs,P.N.Belhumeur and R.Basri,”Comparing Images under Variable Illumination Proceedings of IEEE Conf.Computer Visual and Pattern Recognition,pp610-617,1998.
[30] A.Shashua and T.Riklin-Raviv,”The Quotient Images: Class-based Re-Rendering and Recognition with Varying Illumination,”IEEE Trans.Pattern Analysis and Machine Intelligence,vol.23,no.2,pp129-139,2001.
[31] Xudong Xie, Kin-Man Lam. Face recognition under varying illumination based on a 2D face shape model [J]. Pattern Recognition 38 ,221-230.2005.
[32]倪伟,郭宝龙,杨镠。“图像多尺度几何分析新进展:Contourlet”。计算机科学,33(2):234-236, 2006.
[33] Rudi,Osher S and Fatemi E.”Nonlinear tota variation based noise removal algorithms“,[J].PhysieaD, 60:259~268, 1992.
[34] S.Osher,A.Marquina,”Explicit algorithms for a new time dependent model based on level set motion for nonlinear deblurring and noise removal”,SIAM J.Sci.Comput.22,pp387-405,2000.
[35] C.R.Vogel,M.E .Oman,”Iterative methods for total variation denoising”,SIAM J.Sci.Stat.Comput.17,pp227-238,1996.
[36] T.F.Chan,P.Mulet.”On the Convergence of the Lagged Diffusivity Fixed Point Method in Total Variation Image Restoration”, SIAM J.Number.Analysis,36,354-367,1999.
[37] T.F.Chan,G.H.Golub&P.Mulet.”A nonlinear primal dual method for total variation based image restoration”, SIAM J.Sci.Comput.20,pp1964-1977,1999.
[38] D.Goldfarb & W.Yin.”Second-Order cone Programming methods for tatal variation-based image restoration”, SIAM J.Sci.Stat.Comput.27,pp662-645,2005.
[39] J.Darbon & M.Sigelle .”Exact Optimization of discrete constrained total variation minization problems”.In:R.Klette and J.Zunic,editors,Tenth International Workshop on Combinatorial Image Analysis,volume 3322 of LNCS,548-557,2004.
[40] J.Darbon & M.Sigelle.”A fast and exact algorithm for total variation minization”.In:J.S.Marques,N.Prez de la Blanca,and P.Pina,editors,2nd Iberian Conference on Pattern Recognition and Image Analysis,volume 3522 of LNCS 351-35,2005.
[41] Y.Wang,J.Yang,W.Yin and Y.zhang,”A New Alternating Minimization Algorithm for Total Variation Image Reconstruction”,UCLA CAM Report 08-32,2008.
[42] MeyerY.”Oscil1ating patterns in image processing and nonlinear evolution equations[R]”.university Lecture Series, American Mathematical.Society,2002.
[43] Vese L.A and Osher.S.”Modeling textures with total variation minimization and Oscil1ating patterns in image processing”.Journal of Scientific Computing,19:553-572,2003.
[44] Chan T.F,Esedoglu S,Aspects of total variation regularized L1 function approximation,SIAM J.Appl.Math.65(5),pp1817-1837,2005.
[45]刘鸣,青岛大学,“基于PDE的图像分解方法研究与应用”,硕士论文,2008。
[46]倪敏,冯承天,上海师范大学理工信息学院,“变分法的Euler-Lagrange方程及其应用”,上海师范大学学报,2000.
[47]李盛武,中山大学,“基于图像高频信息提取与光照归一化的人脸识别研究”,硕士论文,2008.
[48]郑红,青岛大学“,变分图像扩散TV_L1模型的分裂计算方法”,硕士论文,2009.
[49] Terrence Chen,Wotao Yin,Xiang Sean Zhou,“Total Variation Models for Variable Lighting Face Recognition”, IEEE Transactions on Pattern analysis and machine international,vol.28,no.9,2006.
[50] P.J.Burt,E.H.Adelson.The Laplacian pyramid as a compact image code,IEEE Trans.Communication,pp532-540.1983.
[51] R.H.Bamberger,M.J.Smith,Afilter bank for the directionl decomposition of images: Theory and design,IEEE Trans.Signal Processing,pp882-893.1992.
[52]周晓,朱才志,“偏微分方程在图像处理中的应用”,安徽教育学院学报,Vol.25 NO.3,2007.
[53]庄连生,中国科学技术大学,“复杂光照条件下人脸识别关键算法研究”,博士论文,2006.
[2] N. Furl, A. J. O’Toole, and P. J. Phillips,“Face recognition algorithms as models of the other race effect,”Cogn. Sci., vol. 96, pp. 1–19, 2002.
[3] A. Adler and J. Maclean,“Performance comparison of human and automatic face recognition,”in Proc. Biometrics Consortium Conf., Washington, DC, Sep. 20–22, 2004.
[4] C. H. Liu, H. Seetzen, A. M. Burton, and A. Chaudhuri,“Face recognition is robust with incongruent image resolution: Relationship to security video images,”J. Exper. Psychol.,Appl., vol. 9, no. 1, pp. 33–41, Mar. 2003.
[5] R.J. Baron. Mechanisms of Human Facial Recognition. Int.J. Man-Machine Studies,1981, 15, pp:137-178.
[6] L. Wiskott, J.M. Fellous and N. Kruger. Face Recognition by Elastic Bunch Graph Matching. IEEE Trans. Pattern Analysis and Machine Intelligence, 19(7),pp:775-779. 1997.
[7] M. Turk and A. Pentland, "Eigenfaces for Recognition", Journal of cognitive neuroscience, 1991 3(1), pp:71-86.
[8] B.Moghaddam.”Principal manifolds and probabilistic subspaces for visual recognition’,IEEE Trans.PAMI 24(6) 780-788.2002.
[9] B.Moghaddam and A. Pentland.”Probabilistic Visual Learning for Object Detection”.Proc.Int’l Conf.Computer Vision,pp.786-793,1995.
[10] B.Moghaddam and A. Pentland.”Probabilistic Visual Learning for Object Representation”. IEEE Trans.PAMI,vol.20,n0.7,pp.696-710,1997.
[11] Baback Moghaddam,Tony Jebara and Alex Pentland.”Bayesian Face Recognition’.Pattern Recognition Vol.33(2000),pp1771-1782,2000.
[12] P.Belhumeur,J.Hespanha and D.Kriegman.”Eigenfaces vs.Fisherfaces:Recogition using class specific linear projection”. in Processing of Fourth European Conference on Computer Vision ECCV’96,pp45-56,1996.
[13] P.N. Belhumeur, J.Hespanha and D.Kriegman.”Eigenfaces vs.Fisherfaces:Recogition using class specific linear projection”.IEEE Transacitions on Pattern Analysis and Machine Intelligence,Special Issue on Face Recognition,17(7):711-720,1997.
[14] R.Duda,P.Hart,D.stork,”Pattern Classidication”,John Wiley & Sons,NewYork,2001.
[15] H.Yu and H.Yang,”A direct LDA algorithm for high-dimensional data-with application to face recognition’,Pattern Recognition,vol.34,no.10,pp2067-2070,2001.
[16] C.Liu and H.Wechsler,”Enhanced Fisher Linear Discriminant Models for Face Recognition,”Proceeedings of International Conference on Pattern Recognition,vol.2,pp.1368-1372,1998.
[17] L.F.Chen.H.M.Liao,J.C.Lin,M.T.Ko,and G.J.Yu.”Anew LDA-based Face recognition Systerm Which Can Solve the Small Sample Size Problem’.Pattern Recognition,vol.33,no.10,pp1712-1726,2000.
[18] M.H.Yang,N,Ahuja,and D.Kriegman.”Face Recognition Using Kerne Eigenfaces’,Int. Conf on image Processing ,vol.1,pp37-40.,2000.
[19] M.H.Yang.”Kernel Eigenfaces vs Kernel Fisherfaces:Face Recognition Using Kernel Methods’,Proc.Int’l,Conf.Automatic Face Gesture Recognition,pp.215-220,2002.
[20] M.Propp and A.Samal,”Artificial Neural Network Architectures for Face Detection,”Intelligent Eng.Systerms through Artificial Neural Network,vol.2,1992.
[21]A.N.Pajagopalan,K.S.Kumar,J.Karlekar,R.Manivasakan,M.M.Patil,U.B.Desai, P.G.Poonacha,and S.Chaudburi,”Finding Faces in Photographs,”Proceeding of IEEE Conf. Computer Vision,pp640-645,1998.
[22] P.N. Belhumeur and D.J.Kriegman,”what is the set of images of an Object under all Possible Lighting Conditions”IEEE Int’l Conf Computer Vision and Pattern Recognization,2004.
[23] R.Basri and D.Jacobs,”Lambertain Refletance and Linear Subspaces,”Technical Report 2000-172R,NEC Research Inst,2000.
[24] K.K.Sung and T.Poggio,”Example Based Learning for View-Based Human Face Detection,”IEEE Trans.Pattern Analysis and Machine Intelligence,vol.20,no.1,pp39-51,1998.
[25] B.Moghaddam and A.Pentland,”Probabilistic Visual Learning for Object Representation,”IEEE Trans.Pattern Analysis and Machine Intelligence,vol.19.no.17,pp696-710,1997.
[26] R.Brunelli and T.Poggio,”Hyper BF Networks for Real Object Recognition,”Praceedings of International Joint Conf.on Artificial Intelligence,pp1278-1284,1991.
[27] V.Govindaraju,”Locating human faces in photographs,”International Journal of Computer Vision,vol.19,no.2.pp129-146,1996.
[28] A.Shashua,”Geometry and Photometry in 3D Visual Recognition.”Ph.D.MIT,Artificial Intelligence Lab,1992.
[29] D.Jacobs,P.N.Belhumeur and R.Basri,”Comparing Images under Variable Illumination Proceedings of IEEE Conf.Computer Visual and Pattern Recognition,pp610-617,1998.
[30] A.Shashua and T.Riklin-Raviv,”The Quotient Images: Class-based Re-Rendering and Recognition with Varying Illumination,”IEEE Trans.Pattern Analysis and Machine Intelligence,vol.23,no.2,pp129-139,2001.
[31] Xudong Xie, Kin-Man Lam. Face recognition under varying illumination based on a 2D face shape model [J]. Pattern Recognition 38 ,221-230.2005.
[32]倪伟,郭宝龙,杨镠。“图像多尺度几何分析新进展:Contourlet”。计算机科学,33(2):234-236, 2006.
[33] Rudi,Osher S and Fatemi E.”Nonlinear tota variation based noise removal algorithms“,[J].PhysieaD, 60:259~268, 1992.
[34] S.Osher,A.Marquina,”Explicit algorithms for a new time dependent model based on level set motion for nonlinear deblurring and noise removal”,SIAM J.Sci.Comput.22,pp387-405,2000.
[35] C.R.Vogel,M.E .Oman,”Iterative methods for total variation denoising”,SIAM J.Sci.Stat.Comput.17,pp227-238,1996.
[36] T.F.Chan,P.Mulet.”On the Convergence of the Lagged Diffusivity Fixed Point Method in Total Variation Image Restoration”, SIAM J.Number.Analysis,36,354-367,1999.
[37] T.F.Chan,G.H.Golub&P.Mulet.”A nonlinear primal dual method for total variation based image restoration”, SIAM J.Sci.Comput.20,pp1964-1977,1999.
[38] D.Goldfarb & W.Yin.”Second-Order cone Programming methods for tatal variation-based image restoration”, SIAM J.Sci.Stat.Comput.27,pp662-645,2005.
[39] J.Darbon & M.Sigelle .”Exact Optimization of discrete constrained total variation minization problems”.In:R.Klette and J.Zunic,editors,Tenth International Workshop on Combinatorial Image Analysis,volume 3322 of LNCS,548-557,2004.
[40] J.Darbon & M.Sigelle.”A fast and exact algorithm for total variation minization”.In:J.S.Marques,N.Prez de la Blanca,and P.Pina,editors,2nd Iberian Conference on Pattern Recognition and Image Analysis,volume 3522 of LNCS 351-35,2005.
[41] Y.Wang,J.Yang,W.Yin and Y.zhang,”A New Alternating Minimization Algorithm for Total Variation Image Reconstruction”,UCLA CAM Report 08-32,2008.
[42] MeyerY.”Oscil1ating patterns in image processing and nonlinear evolution equations[R]”.university Lecture Series, American Mathematical.Society,2002.
[43] Vese L.A and Osher.S.”Modeling textures with total variation minimization and Oscil1ating patterns in image processing”.Journal of Scientific Computing,19:553-572,2003.
[44] Chan T.F,Esedoglu S,Aspects of total variation regularized L1 function approximation,SIAM J.Appl.Math.65(5),pp1817-1837,2005.
[45]刘鸣,青岛大学,“基于PDE的图像分解方法研究与应用”,硕士论文,2008。
[46]倪敏,冯承天,上海师范大学理工信息学院,“变分法的Euler-Lagrange方程及其应用”,上海师范大学学报,2000.
[47]李盛武,中山大学,“基于图像高频信息提取与光照归一化的人脸识别研究”,硕士论文,2008.
[48]郑红,青岛大学“,变分图像扩散TV_L1模型的分裂计算方法”,硕士论文,2009.
[49] Terrence Chen,Wotao Yin,Xiang Sean Zhou,“Total Variation Models for Variable Lighting Face Recognition”, IEEE Transactions on Pattern analysis and machine international,vol.28,no.9,2006.
[50] P.J.Burt,E.H.Adelson.The Laplacian pyramid as a compact image code,IEEE Trans.Communication,pp532-540.1983.
[51] R.H.Bamberger,M.J.Smith,Afilter bank for the directionl decomposition of images: Theory and design,IEEE Trans.Signal Processing,pp882-893.1992.
[52]周晓,朱才志,“偏微分方程在图像处理中的应用”,安徽教育学院学报,Vol.25 NO.3,2007.
[53]庄连生,中国科学技术大学,“复杂光照条件下人脸识别关键算法研究”,博士论文,2006.