基于近红外图像的实时高性能人脸识别算法的研究
|
摘要
近年来,生物特征识别受到了人们越来越多的关注。与其他生物特征识别方法相比,人脸识别具有直接、友好、方便等特点,易于为用户所接受,因而人脸识别技术在社会各领域都得到了广泛的研究与应用。
在人脸识别中,光照变化与姿态变化通常会引起人脸外观的明显变化。目前,解决光照问题的一种有效方法是利用主动近红外光进行人脸识别,该方法可以有效的减少光照变化对识别性能的影响。然而,人脸姿态变化很难通过直接的方法进行控制,因此可以采用姿态估计来进一步克服姿态变化问题。
人脸姿态估计是指在输入的图像中确定人脸在三维空间中姿态的过程,人脸姿态变化通常分成3种类型：左右旋转、上下旋转和图像平面内旋转。
本文提出了一种新的人脸姿态估计方法,用于计算人脸左右旋转角度。该方法首先对归一化人脸图像进行鼻孔定位,再根据模型中鼻孔与眼睛坐标参数计算得到人脸的左右旋转角度。由于人脸左右旋转角度过大时,易导致一个鼻孔不可见,此时鼻孔定位算法失效。在此情况下,采用边缘统计方法进行人脸姿态估计。边缘统计方法首先对归一化人脸图像进行边缘检测,再对边缘图像对称展开,对展开后的图像做垂直直方图统计,计算直方图平均值与物理中线的水平距离以确定人脸的左右旋转角度。
本文的人脸识别采用LBP特征和Ada Boost算法进行人脸识别分类器训练。在识别过程中,首先对待识别图像进行人脸姿态估计,若人脸左右旋转角度过大,则认为此图像质量较低,不再进行后续的人脸识别处理。人脸姿态估计预处理可以确保所有参与识别的图像具有正脸的特点,这有利于提高人脸识别系统的性能。
本文实现了基于LBP特征和Ada Boost算法的人脸识别训练程序,在一个较大规模的近红外人脸数据库上进行训练。在测试库上,使用人脸识别分类器进行识别。实验结果表明,带有人脸姿态估计预处理的人脸识别方法性能优于没有人脸姿态预处理的人脸识别方法,具有更高的精确性,且达到了实时性要求。
In recent years, biometric identification technology is getting increasingly attention in public security systems. Comparing with other biometric technology, face recognition has some advantages, such as easy to use, harder to circumvent etc. Now, it is widely used in society.
The variations of illumination and pose will affect the performance of the face recognition system seriously. Using active Near-Infrared Imaging system can solve the problem caused by variations of illumination effectively. However, the variation of pose cannot be controlled directly; we can estimate the face pose to solve the problem. The task of face pose estimation is to determine the pose of the input face image in 3-D space.
This thesis proposes a novel face pose estimation method, which is used for determinate rotation angle of a given face. The novel method consists of edge statistical-based face pose estimation process and feature location-based face pose estimation process. The edge statistical-based face pose estimation process uses the overall information to extract the features firstly, and then the linear regression algorithm is applied to build the corresponding relationship between features and poses. Feature location-based face pose estimation process is another process proposed in this thesis, which is a local information-based method utilizing mathematical morphology operator to pinpoint the nostril. Then according to the location of the nose and eyes, face poses are estimated.
In this thesis, LBP feature and AdaBoost algorithm is used to train a classifier, which can discriminate the label of the input pattern. In the process of face recognition, the input face image was estimated by the method mentioned above. If the rotation angle of the face exceeds the threshold, the image is discarded. The pre-judgment process improves performance of the face recognition system.
The trained classifier was tested in a large scale database, experimental results shows that the face recognition method with pre-judgment of pose estimation outperforms traditional method.
在人脸识别中,光照变化与姿态变化通常会引起人脸外观的明显变化。目前,解决光照问题的一种有效方法是利用主动近红外光进行人脸识别,该方法可以有效的减少光照变化对识别性能的影响。然而,人脸姿态变化很难通过直接的方法进行控制,因此可以采用姿态估计来进一步克服姿态变化问题。
人脸姿态估计是指在输入的图像中确定人脸在三维空间中姿态的过程,人脸姿态变化通常分成3种类型：左右旋转、上下旋转和图像平面内旋转。
本文提出了一种新的人脸姿态估计方法,用于计算人脸左右旋转角度。该方法首先对归一化人脸图像进行鼻孔定位,再根据模型中鼻孔与眼睛坐标参数计算得到人脸的左右旋转角度。由于人脸左右旋转角度过大时,易导致一个鼻孔不可见,此时鼻孔定位算法失效。在此情况下,采用边缘统计方法进行人脸姿态估计。边缘统计方法首先对归一化人脸图像进行边缘检测,再对边缘图像对称展开,对展开后的图像做垂直直方图统计,计算直方图平均值与物理中线的水平距离以确定人脸的左右旋转角度。
本文的人脸识别采用LBP特征和Ada Boost算法进行人脸识别分类器训练。在识别过程中,首先对待识别图像进行人脸姿态估计,若人脸左右旋转角度过大,则认为此图像质量较低,不再进行后续的人脸识别处理。人脸姿态估计预处理可以确保所有参与识别的图像具有正脸的特点,这有利于提高人脸识别系统的性能。
本文实现了基于LBP特征和Ada Boost算法的人脸识别训练程序,在一个较大规模的近红外人脸数据库上进行训练。在测试库上,使用人脸识别分类器进行识别。实验结果表明,带有人脸姿态估计预处理的人脸识别方法性能优于没有人脸姿态预处理的人脸识别方法,具有更高的精确性,且达到了实时性要求。
In recent years, biometric identification technology is getting increasingly attention in public security systems. Comparing with other biometric technology, face recognition has some advantages, such as easy to use, harder to circumvent etc. Now, it is widely used in society.
The variations of illumination and pose will affect the performance of the face recognition system seriously. Using active Near-Infrared Imaging system can solve the problem caused by variations of illumination effectively. However, the variation of pose cannot be controlled directly; we can estimate the face pose to solve the problem. The task of face pose estimation is to determine the pose of the input face image in 3-D space.
This thesis proposes a novel face pose estimation method, which is used for determinate rotation angle of a given face. The novel method consists of edge statistical-based face pose estimation process and feature location-based face pose estimation process. The edge statistical-based face pose estimation process uses the overall information to extract the features firstly, and then the linear regression algorithm is applied to build the corresponding relationship between features and poses. Feature location-based face pose estimation process is another process proposed in this thesis, which is a local information-based method utilizing mathematical morphology operator to pinpoint the nostril. Then according to the location of the nose and eyes, face poses are estimated.
In this thesis, LBP feature and AdaBoost algorithm is used to train a classifier, which can discriminate the label of the input pattern. In the process of face recognition, the input face image was estimated by the method mentioned above. If the rotation angle of the face exceeds the threshold, the image is discarded. The pre-judgment process improves performance of the face recognition system.
The trained classifier was tested in a large scale database, experimental results shows that the face recognition method with pre-judgment of pose estimation outperforms traditional method.
引文
[1]Bledose W. Man-machine facial recognition. Panoranic Research Inc, Palo Alto, CA,1966, Rep PR I:22.
[2]Marr. A computational investigation into the human representation and processing of visual information.1982.
[3]Lades M, Vorbueggen J C, Buhmann J, et al. Distortion invariant object recognition in the dynamic link architecture. IEEE Transactions on Computers.1991,42(3):300-311.
[4]Wiskott L, Fellous J M, Kruger N, et al. Face recognition by elastic bunch graph matching. IEEE Transactions on Pattern Analysis and Machine Intelligence.1997,19(7):775-779.
[5]Kirby M, SIrovich L. Application of Karhunen-Loeve procedure for the characterization of human faces. IEEE Transactions on Pattern Analysis and Machine Intelligence.1990, 12(1):102-108.
[6]Turk M, Pentland A. Eigenfaces for recognition. Journal of Cognitive Neuroscience.1991, 3(1):71-86.
[7]Fisher R A. The use of multiple measurements in taxonomic problems. Annals of Eugenics. 1936,7:179-188.
[8]Rao C R. The utilization of multiple measurements in problems of biological classification. Journal of the Royal Statistical Society series B.1948,10:159-203.
[9]Cottes T F, Taylor C J, Cooper D H, et al. Active shape models-Their training and application. Computer Vision and Image Understanding.1995,61(1):38-59.
[10]Cootes T F, Edwards G J, Taylor C J. Active appearance models. IEEE Transactions on Pattern Analysis and Machine Intelligence.2001,23(6):681-685.
[11]Liu C and Wechsler H. Gabor Feature Based Classification Using the Enhanced Fisher Linear Discriminant Model for Face Recognition. IEEE Trans. Image Processing.2002, 11(4):467-476.
[12]Shan S, Gao W, Chang Y, Cao B, Yang P. Review the Strength of Gabor features for face recognition from the angle of its robustness to mis-alignment. Proceedings of ICPR2004. 2004,1:338-341.
[13]Lee T S. Image Representation Using 2d Gabor Wavelets. IEEE Trans. Pattern Analysis and Machine Intelligence.1996,18(10):959-971.
[14]Ojala T, Pietikainen M, Harwood D. A Comparative Study of Texture Measure with Classification based on Feature distributions. Pattern Recognition.1996,29:51-59.
[15]Li SZ, Chu R, Liao S, Zhang L. Illumination invariant face recognition using near-infrared images. IEEE Transactions on Pattern Analysis and Machine Intelligence,2007,29(4): 627-639.
[16]Georghiades AS, Belhumeur PN, Kriegman DJ. From Few to Many:Illumination Cone Models for Face Recognition under Variable Lighting and Pose. IEEE Transactions on Pattern Analysis and Machine Intelligence,2001,23(6):643-660.
[17]梁国远,查红彬,刘宏.基于三维模型和仿射对应原理的人脸姿态估计方法.计算机学报,2005,28(5)：792-800.
[18]Ji Q, Hu R.3D face pose estimation and tracking from a monocular camera. Image and Vision Computing,2002,20(7):499-511.
[19]Jeffrey Ng, Gong SG. Composite support vector machines for detection of faces across views and pose estimation. Image and Vision Computing,2002,20 (5/6):359-368.
[20]Kriiger N, Potzsch M, Malsburg CVD. Determination of face position and pose with a learned representation based on labeled graphs. Image and Vision Computing,1997, 15(8):665-673.
[21]Li SZ, Lu XG, Hou XW, Peng XH, Cheng QS. Learning multiview face subspaces and facial pose estimation using independent component analysis. IEEE Trans. on Image Processing,2005,14(6):705-712.
[22]Raytchev B, Yoda I, Sakaue K. Head pose estimation by nonlinear manifold learning. In: Goldfarb L, ed. Proc. of the Int'l Conf. on Pattern Recognition, Washington:IEEE Computer Society,2004,4:462-466.
[23]刘淼,郭东伟,马捷,孙浩翔,周春光.基于椭圆模型和神经网络的人脸姿态估计方法.吉林大学学报(理学版),2008,46(4)：687-690.
[24]McCulloch W S, Pitts W H. A logical calculus of the ideas immanent in nervous activity. Bulletion of Mathematical Biophysics.1943,5:115-133.
[25]Vapnik V N. Statistical Learning Theory.Wiley-Interscience.1998.
[26]Lam K, Hong Y. An Analytic-to-Holistic Approach for Face Recognition Based on a Single Frontal View. IEEE Transactions on Pattern Analysis and Machine Intelligence.1998, 20(7):673-686.
[27]Yegnanarayana B, Anil Kumar sao, Vijaya Kumar BVK, Marios Savvides. Determination of pose angle of face using dynamic space warping. Proceedings of the International Conference on Information Technology:Coding and Computing (ITCC'04),2004, 1(5/7):661-664.
[28]王基帆,童卫青.基于数理形态学的近红外光图像实时人脸检测.华东师范大学学报(自然科学版),2010,3：39-47.
[29]Timo A, Abdenour H, Matti P. Face Recognition with local binary patterns. Proc. Of the European Conf. on Computer Vision. LNCS 3021, Prague:Springer-Verlag,.2004:469-481.
[30]Ojala T, Pietikainen M, Maenpaa T. Multiresolution gray-scale and rotation invariant texture classification with local binary patterns. IEEE Transactions on Pattern Analysis and Machine Intelligence.2002,24:971-987.
[31]Valiant L. A theory of the learnable. Communications of the ACM,1984, 27(11):1134-1142.
[32]Schapire R E. The strength of weak learnability. Machine Learning.1990,5:197-227.
[33]Freund Y, Schapire R E. A decision-theoretic generalization of online learning and an application to boosting. In Proc. Second European Conference on Computational Learning Theory,1994.
[34]Schapire R E. A brief introduction to boosting. Proceedings of the 16th International Joint Conference on Artificial Intelligence.1999.
[35]Freund Y, Schapire R E. Experiments with a new boosting algorithm. Machine Learning: Proceedings of the 13th International Conference. Morgan Kauffman,1996:148-156.
[36]Viola P, Jones M. Rapid objects detection using a boosted cascade of simple features. Proceeding of 2001 IEEE Computer Vision and Pattern Recognition.2001,1:511-518.
[37]Baback Moghaddam, Tony Jebara, Alex Pentland. Bayesian Face Recognition. Pattern Recognition.2000,33:1771-1782.
[38]http://www.hongeltec.cn
[2]Marr. A computational investigation into the human representation and processing of visual information.1982.
[3]Lades M, Vorbueggen J C, Buhmann J, et al. Distortion invariant object recognition in the dynamic link architecture. IEEE Transactions on Computers.1991,42(3):300-311.
[4]Wiskott L, Fellous J M, Kruger N, et al. Face recognition by elastic bunch graph matching. IEEE Transactions on Pattern Analysis and Machine Intelligence.1997,19(7):775-779.
[5]Kirby M, SIrovich L. Application of Karhunen-Loeve procedure for the characterization of human faces. IEEE Transactions on Pattern Analysis and Machine Intelligence.1990, 12(1):102-108.
[6]Turk M, Pentland A. Eigenfaces for recognition. Journal of Cognitive Neuroscience.1991, 3(1):71-86.
[7]Fisher R A. The use of multiple measurements in taxonomic problems. Annals of Eugenics. 1936,7:179-188.
[8]Rao C R. The utilization of multiple measurements in problems of biological classification. Journal of the Royal Statistical Society series B.1948,10:159-203.
[9]Cottes T F, Taylor C J, Cooper D H, et al. Active shape models-Their training and application. Computer Vision and Image Understanding.1995,61(1):38-59.
[10]Cootes T F, Edwards G J, Taylor C J. Active appearance models. IEEE Transactions on Pattern Analysis and Machine Intelligence.2001,23(6):681-685.
[11]Liu C and Wechsler H. Gabor Feature Based Classification Using the Enhanced Fisher Linear Discriminant Model for Face Recognition. IEEE Trans. Image Processing.2002, 11(4):467-476.
[12]Shan S, Gao W, Chang Y, Cao B, Yang P. Review the Strength of Gabor features for face recognition from the angle of its robustness to mis-alignment. Proceedings of ICPR2004. 2004,1:338-341.
[13]Lee T S. Image Representation Using 2d Gabor Wavelets. IEEE Trans. Pattern Analysis and Machine Intelligence.1996,18(10):959-971.
[14]Ojala T, Pietikainen M, Harwood D. A Comparative Study of Texture Measure with Classification based on Feature distributions. Pattern Recognition.1996,29:51-59.
[15]Li SZ, Chu R, Liao S, Zhang L. Illumination invariant face recognition using near-infrared images. IEEE Transactions on Pattern Analysis and Machine Intelligence,2007,29(4): 627-639.
[16]Georghiades AS, Belhumeur PN, Kriegman DJ. From Few to Many:Illumination Cone Models for Face Recognition under Variable Lighting and Pose. IEEE Transactions on Pattern Analysis and Machine Intelligence,2001,23(6):643-660.
[17]梁国远,查红彬,刘宏.基于三维模型和仿射对应原理的人脸姿态估计方法.计算机学报,2005,28(5)：792-800.
[18]Ji Q, Hu R.3D face pose estimation and tracking from a monocular camera. Image and Vision Computing,2002,20(7):499-511.
[19]Jeffrey Ng, Gong SG. Composite support vector machines for detection of faces across views and pose estimation. Image and Vision Computing,2002,20 (5/6):359-368.
[20]Kriiger N, Potzsch M, Malsburg CVD. Determination of face position and pose with a learned representation based on labeled graphs. Image and Vision Computing,1997, 15(8):665-673.
[21]Li SZ, Lu XG, Hou XW, Peng XH, Cheng QS. Learning multiview face subspaces and facial pose estimation using independent component analysis. IEEE Trans. on Image Processing,2005,14(6):705-712.
[22]Raytchev B, Yoda I, Sakaue K. Head pose estimation by nonlinear manifold learning. In: Goldfarb L, ed. Proc. of the Int'l Conf. on Pattern Recognition, Washington:IEEE Computer Society,2004,4:462-466.
[23]刘淼,郭东伟,马捷,孙浩翔,周春光.基于椭圆模型和神经网络的人脸姿态估计方法.吉林大学学报(理学版),2008,46(4)：687-690.
[24]McCulloch W S, Pitts W H. A logical calculus of the ideas immanent in nervous activity. Bulletion of Mathematical Biophysics.1943,5:115-133.
[25]Vapnik V N. Statistical Learning Theory.Wiley-Interscience.1998.
[26]Lam K, Hong Y. An Analytic-to-Holistic Approach for Face Recognition Based on a Single Frontal View. IEEE Transactions on Pattern Analysis and Machine Intelligence.1998, 20(7):673-686.
[27]Yegnanarayana B, Anil Kumar sao, Vijaya Kumar BVK, Marios Savvides. Determination of pose angle of face using dynamic space warping. Proceedings of the International Conference on Information Technology:Coding and Computing (ITCC'04),2004, 1(5/7):661-664.
[28]王基帆,童卫青.基于数理形态学的近红外光图像实时人脸检测.华东师范大学学报(自然科学版),2010,3：39-47.
[29]Timo A, Abdenour H, Matti P. Face Recognition with local binary patterns. Proc. Of the European Conf. on Computer Vision. LNCS 3021, Prague:Springer-Verlag,.2004:469-481.
[30]Ojala T, Pietikainen M, Maenpaa T. Multiresolution gray-scale and rotation invariant texture classification with local binary patterns. IEEE Transactions on Pattern Analysis and Machine Intelligence.2002,24:971-987.
[31]Valiant L. A theory of the learnable. Communications of the ACM,1984, 27(11):1134-1142.
[32]Schapire R E. The strength of weak learnability. Machine Learning.1990,5:197-227.
[33]Freund Y, Schapire R E. A decision-theoretic generalization of online learning and an application to boosting. In Proc. Second European Conference on Computational Learning Theory,1994.
[34]Schapire R E. A brief introduction to boosting. Proceedings of the 16th International Joint Conference on Artificial Intelligence.1999.
[35]Freund Y, Schapire R E. Experiments with a new boosting algorithm. Machine Learning: Proceedings of the 13th International Conference. Morgan Kauffman,1996:148-156.
[36]Viola P, Jones M. Rapid objects detection using a boosted cascade of simple features. Proceeding of 2001 IEEE Computer Vision and Pattern Recognition.2001,1:511-518.
[37]Baback Moghaddam, Tony Jebara, Alex Pentland. Bayesian Face Recognition. Pattern Recognition.2000,33:1771-1782.
[38]http://www.hongeltec.cn