基于特征脸的面部识别技术研究
|
摘要
面部识别也称为人脸识别,是生物特征识别技术的一个主要方向,是应用最广泛的生物识别方法技术之一。与其他生物特征相比,人脸识别具有主动性、非侵犯性和用户友好性等特点。人脸识别是模式识别技术在图像领域中的具体运用,其应用前景非常广阔,可以应用到身份证件的鉴别、自动门禁控制系统、刑侦破案、视频监控、家庭安全等领域。
人脸自动识别系统包括人脸图像预处理、特征值提取、人脸识别三个部分。
人脸图像预处理是人脸识别过程中的一个重要的步骤。输入图像从现实生活中的景物转换成数字图像信息时,由于摄像设备条件不一样,如设备性能的优劣以及照相时光照明暗程度不同等,往往存在噪声、光照变化等缺陷,因此必须在提取特征之前对图像作预处理的工作。本文使用了彩色图像灰度化,中值滤波,灰度归一化,定位人脸,几何缩放等技术来消除这些影响,提高了识别率。
本文采取了基于小波变换和离散余弦变换相结合的特征值提取方法。图像经过小波变换以后,图像的能量和绝大部分信息保留在了低频部分,同时经过小波变换后,在图像的敏感位置生成的特征矢量模相对较大。离散余弦变换是一种正交变换,我们可以从数学知识上知道各种正交变换都有能够减小随机向量相关性的特性。信号经过多数正交变换以后,系数分布比较集中,且能量集中在少数的变换系数上。这些优点应用于人脸图像,有利于人脸的识别。
本文首先对输入的图像进行一系列预处理,然后利用小波变换对经过预处理后的人脸图像做两次小波分解,再通过离散余弦变换对低频分量作进一步的特征提取和压缩,提取100个离散余弦变换系数作为最后的特征值。最后利用欧氏距离和最近邻分类器进行识别。
基于以上理论,可以把人脸识别技术应用于很多方面,比如说应用在门禁系统。对给定的人脸图像先进行预处理,然后提取特征值,最后与己知人脸库中存储的模型进行匹配比较,确定是否是库中某一人物,如果是,则开启门禁系统,实现自动识别的目的。本文从实验的角度做了初步的测试,结果证明该方法确实可行。
Face recognition is a major direction of biology recognition technology. It is one of the most extensively used technologies. Compared with other biometrics, face recognition has features such as initiative, non-invasive and user-friendliness.Face recognition is the specific use of pattern recognition in the image field. It has broad application prospect, such as identification of identity documents, automated access control system, criminal detection, video surveillance, home security and other fields.
Face recognition system includes face image preprocessing, feature extraction and face recognition three parts.
Face image preprocessing is an important step for face recognition process. When the input image from the actual scene into digital image information, there are often the defect such as existence of noise and illumination changes due to equipment conditions, such as the degree of illumination bright-dark, as well as the merits of performance. So image preprocessing is a necessary work before the extraction of characteristics. Color image gray processing, median filter, histogram equalization,face location and geometric scaling are used to eliminate these effects and improve the recognition rate.
The feature extraction based on the combining of wavelet transform and discrete cosine transform is used in the paper. Image is decomposed using wavelet transform and its low-frequency part reserves the majority of information and energy. At the same time, the relatively larger feature vector modulus is generated in the sensitive location of the image after the wavelet transform. Discrete cosine transform is an orthogonal transformation. A variety of orthogonal transformation can reduce the relevance of random vector in a certain extent and the energy will be concentrated on a small number of transform coefficients when the signal was transformed by most of the orthogonal transformation. This can be proved in the mathematics. Those are useful for face recognition when these advantages were used in face image. In this algorithm, first of all, face image after preprocessed is decomposed by wavelet transform twice, and then the low-frequency components are transformed by discrete cosine transform to extract the feature and are compressed. 100 discrete cosine transform coefficients are extracted to be the last feature values. Finally, euclidean distance and nearest neighbor classifier are used to recognizing target.
Based on above theory, the face recognition technology coule be applied in many fields. For example: the access control system. First, we can preprocess the given face image. Second, feature value is extracted. At last, comparing with face image and the model in the face database to determine whether it belongs to the face database. If so, access control system will be opened and this can realize the purpose of automatic recognition. The results of experiment indicate the new method of the technology of human face recognition is available.
人脸自动识别系统包括人脸图像预处理、特征值提取、人脸识别三个部分。
人脸图像预处理是人脸识别过程中的一个重要的步骤。输入图像从现实生活中的景物转换成数字图像信息时,由于摄像设备条件不一样,如设备性能的优劣以及照相时光照明暗程度不同等,往往存在噪声、光照变化等缺陷,因此必须在提取特征之前对图像作预处理的工作。本文使用了彩色图像灰度化,中值滤波,灰度归一化,定位人脸,几何缩放等技术来消除这些影响,提高了识别率。
本文采取了基于小波变换和离散余弦变换相结合的特征值提取方法。图像经过小波变换以后,图像的能量和绝大部分信息保留在了低频部分,同时经过小波变换后,在图像的敏感位置生成的特征矢量模相对较大。离散余弦变换是一种正交变换,我们可以从数学知识上知道各种正交变换都有能够减小随机向量相关性的特性。信号经过多数正交变换以后,系数分布比较集中,且能量集中在少数的变换系数上。这些优点应用于人脸图像,有利于人脸的识别。
本文首先对输入的图像进行一系列预处理,然后利用小波变换对经过预处理后的人脸图像做两次小波分解,再通过离散余弦变换对低频分量作进一步的特征提取和压缩,提取100个离散余弦变换系数作为最后的特征值。最后利用欧氏距离和最近邻分类器进行识别。
基于以上理论,可以把人脸识别技术应用于很多方面,比如说应用在门禁系统。对给定的人脸图像先进行预处理,然后提取特征值,最后与己知人脸库中存储的模型进行匹配比较,确定是否是库中某一人物,如果是,则开启门禁系统,实现自动识别的目的。本文从实验的角度做了初步的测试,结果证明该方法确实可行。
Face recognition is a major direction of biology recognition technology. It is one of the most extensively used technologies. Compared with other biometrics, face recognition has features such as initiative, non-invasive and user-friendliness.Face recognition is the specific use of pattern recognition in the image field. It has broad application prospect, such as identification of identity documents, automated access control system, criminal detection, video surveillance, home security and other fields.
Face recognition system includes face image preprocessing, feature extraction and face recognition three parts.
Face image preprocessing is an important step for face recognition process. When the input image from the actual scene into digital image information, there are often the defect such as existence of noise and illumination changes due to equipment conditions, such as the degree of illumination bright-dark, as well as the merits of performance. So image preprocessing is a necessary work before the extraction of characteristics. Color image gray processing, median filter, histogram equalization,face location and geometric scaling are used to eliminate these effects and improve the recognition rate.
The feature extraction based on the combining of wavelet transform and discrete cosine transform is used in the paper. Image is decomposed using wavelet transform and its low-frequency part reserves the majority of information and energy. At the same time, the relatively larger feature vector modulus is generated in the sensitive location of the image after the wavelet transform. Discrete cosine transform is an orthogonal transformation. A variety of orthogonal transformation can reduce the relevance of random vector in a certain extent and the energy will be concentrated on a small number of transform coefficients when the signal was transformed by most of the orthogonal transformation. This can be proved in the mathematics. Those are useful for face recognition when these advantages were used in face image. In this algorithm, first of all, face image after preprocessed is decomposed by wavelet transform twice, and then the low-frequency components are transformed by discrete cosine transform to extract the feature and are compressed. 100 discrete cosine transform coefficients are extracted to be the last feature values. Finally, euclidean distance and nearest neighbor classifier are used to recognizing target.
Based on above theory, the face recognition technology coule be applied in many fields. For example: the access control system. First, we can preprocess the given face image. Second, feature value is extracted. At last, comparing with face image and the model in the face database to determine whether it belongs to the face database. If so, access control system will be opened and this can realize the purpose of automatic recognition. The results of experiment indicate the new method of the technology of human face recognition is available.
引文
[1]周杰,卢春雨,张长水,等.人脸自动识别方法综述.电子学报, 2000, (4):102-106
[2] W.BLEDSOE. Man-machine facial recognition. Panoramic Research Inc, Palo Alto, CA, 1966, Rep PRI: 22
[3]张翠平,苏光大.人脸识别技术综述.中国图像图形学报, 2000,5A(11): 885-894
[4] A.L.ALLEN. Personal Descriptions. Lodon: Butterworth, 1950: 17-25
[5] F.I.PARKE. Computer Generated Animation of Faces. Proceedings ACM and Conference, 1972, 1:451-457
[6] R.J.GOLDSTATION, L.D.HARMON, A.B.LESK. Man-machine interactions in human face identification. Bell SystTech Journal, 1972,(51): 399-427
[7]郑坚平,尤婉英.标准正面人脸图像识别.计算机工程. 1992,18(1):1-6
[8] S. JOACHIM. Parallel Genetic Algorithms: Theory and Applications. ISO Press, 1993
[9] A.V.NEFIAN, M.H. Hidden Markov Models for Face Recognition. Internation Conference on Acoustics, Speech and Signal Proceeding, 1998, 99: 2721-2724
[10] D.J.CAVICCHIO. Adaptive Search Using Simulated Evolution. Doctoral Dissertation of University of Michigan, 1970: 25-199
[11] H.CHEN, N.S.FLANN. Parallel Simulated Annealing and Genetic Algorithms: a Space of Hybrid Methods. Berlin: Parallel Problem Solving Form Nature 3. Springer–Verlag, 1994
[12]姜荣,董军宇.人脸识别系统与现实生活中的应用.长春师范学院学报,2006.(1):52-53
[13]舒宁,马洪超,孙和利.模式识别的理论与方法.武汉:武汉大学出版社,2004.(12):83-95
[14]边肇祺,张学工,张长水.模式识别.北京:清华大学出版社, 2000
[15] C.W.CHEN, L.H.HUANG. Human Face Recognition from a Single Front View. Int. Journal of Pattern Recognition ,1992,6(40):571-593
[16]张学工.关于统计学习理论与支持向量机.自动化学报,2000,26(1):32-42
[17] Z.Q.HONG. Algebraic feature extraction of image for recognition. Pattern Recognition, 1991,24(3):211-219
[18]周德龙,高文,赵德斌.基于奇异值分解和判别式KL投影的人脸识别.软件学报, 2003, 14(4): 783-789
[19]王蕴红,谭铁牛,朱勇.基于奇异值分解和数据融合的脸像鉴别.计算机学报,2000,23(6):649-653.
[20]魏小鹏,于万波,金一粟.奇异值方法用于汽车模型识别.中国图像图形学报, 2003, 8A(1):47-50
[21] D.L.ZHOU. A Study of Human Face Recogniton Technology. Xi’an: Northwestern Polytechnical University, 2001
[22]洪子泉,杨静宇.用于图象识别的图象代数特征提取.自动化学报, 1992, 18(2): 233-238
[23] A.TURK, P. PENTLAND. Face Recognition Using Eigenfaces. IEEE Vision and Modeling Group, 1991,5(91):586-591
[24] Hyvarinen. Aapo Fast and robust fixed-point algorithms for independent component analysis, IEEE Transactions on Neural Networks, v 10, n 3, May, 1999, p 626-634
[25] P.C. YUEN, J. HUANG. Face Representation Using Independent Component Analysis. Pattern Recognition, 2002, 35:1247-1257
[26] D.Q. ZHANG, S.C. CHEN, Z.H. ZHOU. A new face recognition method based on SVD perturbation for single example image per person. Applied Mathematics and Computation, 2005,163(2): 895-907
[27]庞彦伟,刘政凯,愈能海,等.融合奇异值分解和线性鉴别分析的人脸识别算法.电路与系统学报, 2006,11(4):47-50
[28] T. ALESANDER, Y. LI, L.J.ZHANG. Exact k-NN queries on clustered SVD datasets. Information Processing Letters, 2005,94(6):247-252
[29] J. FORTUNA, D. CAPSON. Improved support vector classification using PCA and ICA feature space modification. Pattern Recognition, 2007,37(6):1117-1129
[30] J. YANG, X.M. Gao, D. ZHANG, J.Y. YANG. Kernel ICA: An alternative formulation and its application to face recognition. Pattern Recognition, 2005,38(10):1784-1787
[31]蒋宗礼.人工神经网络导论.北京:高等教育出版社, 2002
[32] J. ZHANG, Y. YAN, M.LADES. Face reognition: Eigenfaces, Elastic Matching, and Neural Nets. IEEE Proc. 1998,85(9):1423-1435
[33] F.SAMARIA, S.YOUNG. HMM based architecture for face identification. Image and Computer Vision, 1994,12(4):537-543
[34] L.R. RABINER. A tutorial on hidden markov models and selected application in speech recognition. Proceedings of IEEE, 1989,77(2):257-285
[35] L.WISKOTT, J.M.FELLOUS, N. KRUEGER, etal. Face recognition by elastic bunch graph matching. IEEE Trans, On PAMI, 1997,63(3):665-779
[36] J.Z. MING, J.LIN, X.F. FAN. Feature separation using ICA for a one-dimensional time series and its application in fault detection. Journal of Sound and Vibration, 2005,3(287): 614-624
[37] D. RAMASUBRAMANIAN, Y.V. VENKATESH. Encoding and recognition of faces based on the human visual model and DCT. Pattern Recognition, 2001,34(12):2447-2458
[38]孙永宣,何柯峰,胡良梅.一种新的基于DCT变换的人脸特征.合肥工业大学学报(自然科学版), 2006,29(11):1396-1399
[39] K.ASSALEH, Y.ASSAF. Fearures extraction and analysis for classifying causable patterns in control charts. Computers & Industrial Engineering, 2005,49(1):168-181
[40]赵传强.基于DCT的人脸识别方法研究. [山东大学硕士学位论文]. 2007:40-41
[41]阮秋琦.数字图像处理学.北京:电子工业出版社, 2001.1
[42]杨淑莹. VC++图像处理程序设计.北京:清华大学出版社;北京交通大学出版社,2005.1
[43]周长发.精通Visual C++图像处理编程.北京:电子工业出版社,2000
[44]郎锐.数字图像处理学VC++实现.北京:北京希望电子出版社, 2002
[45]席庆,张春林. Visual C++ 6.0实用编程技术.北京:中国水利水电出版社, 1999
[46]张志军,孙志辉.基于VC平台的彩色图像的灰度化技术.自动化技术与应用.2005,24(1):61-63
[47]李文革.基于主成分分析的人脸识别. [山东大学硕士学位论文],2008: 23-25
[48]任艳斐.直方图均衡化在图像处理中的应用.科技信息, 2007(4):37-38
[49] H.WANG, R.A.HADDAD. Adaptive median filters: New algorithms and results. IEEE Transactions on Image Processing, 1995, 4(4):499-502
[50] I.DAUBESCHIES. Ten Lectures on Wavelets. Philadephia: SIAM Publ, 1993
[51]李世雄.小波变换及其应用.北京:高等教育出版社,1997
[52]马维祯,殷端祥.子波分析与子波变换.广州:华南理工大学出版社, 1996
[53] Z. M.HAFED, M. D.LEVIEN. Face recognition using the discrete cosine transform. Int. J. Comput. Vis, 2001,43(3):167-188
[54] Z.PAN, R.ADAMS, H.BOLOURI. Image redundancy reduction for neural network classification using discrete cosine transforms. In: Proceedings of the IEEE-INNS-ENNS International Joint Conf. On Neural Networks,Como, 2000,Italy 3:149-154.
[55] A.JAIN. Fundamentals of Digital Image Processing. Prentice: Englewood Cliffs,NJ. 1989.
[2] W.BLEDSOE. Man-machine facial recognition. Panoramic Research Inc, Palo Alto, CA, 1966, Rep PRI: 22
[3]张翠平,苏光大.人脸识别技术综述.中国图像图形学报, 2000,5A(11): 885-894
[4] A.L.ALLEN. Personal Descriptions. Lodon: Butterworth, 1950: 17-25
[5] F.I.PARKE. Computer Generated Animation of Faces. Proceedings ACM and Conference, 1972, 1:451-457
[6] R.J.GOLDSTATION, L.D.HARMON, A.B.LESK. Man-machine interactions in human face identification. Bell SystTech Journal, 1972,(51): 399-427
[7]郑坚平,尤婉英.标准正面人脸图像识别.计算机工程. 1992,18(1):1-6
[8] S. JOACHIM. Parallel Genetic Algorithms: Theory and Applications. ISO Press, 1993
[9] A.V.NEFIAN, M.H. Hidden Markov Models for Face Recognition. Internation Conference on Acoustics, Speech and Signal Proceeding, 1998, 99: 2721-2724
[10] D.J.CAVICCHIO. Adaptive Search Using Simulated Evolution. Doctoral Dissertation of University of Michigan, 1970: 25-199
[11] H.CHEN, N.S.FLANN. Parallel Simulated Annealing and Genetic Algorithms: a Space of Hybrid Methods. Berlin: Parallel Problem Solving Form Nature 3. Springer–Verlag, 1994
[12]姜荣,董军宇.人脸识别系统与现实生活中的应用.长春师范学院学报,2006.(1):52-53
[13]舒宁,马洪超,孙和利.模式识别的理论与方法.武汉:武汉大学出版社,2004.(12):83-95
[14]边肇祺,张学工,张长水.模式识别.北京:清华大学出版社, 2000
[15] C.W.CHEN, L.H.HUANG. Human Face Recognition from a Single Front View. Int. Journal of Pattern Recognition ,1992,6(40):571-593
[16]张学工.关于统计学习理论与支持向量机.自动化学报,2000,26(1):32-42
[17] Z.Q.HONG. Algebraic feature extraction of image for recognition. Pattern Recognition, 1991,24(3):211-219
[18]周德龙,高文,赵德斌.基于奇异值分解和判别式KL投影的人脸识别.软件学报, 2003, 14(4): 783-789
[19]王蕴红,谭铁牛,朱勇.基于奇异值分解和数据融合的脸像鉴别.计算机学报,2000,23(6):649-653.
[20]魏小鹏,于万波,金一粟.奇异值方法用于汽车模型识别.中国图像图形学报, 2003, 8A(1):47-50
[21] D.L.ZHOU. A Study of Human Face Recogniton Technology. Xi’an: Northwestern Polytechnical University, 2001
[22]洪子泉,杨静宇.用于图象识别的图象代数特征提取.自动化学报, 1992, 18(2): 233-238
[23] A.TURK, P. PENTLAND. Face Recognition Using Eigenfaces. IEEE Vision and Modeling Group, 1991,5(91):586-591
[24] Hyvarinen. Aapo Fast and robust fixed-point algorithms for independent component analysis, IEEE Transactions on Neural Networks, v 10, n 3, May, 1999, p 626-634
[25] P.C. YUEN, J. HUANG. Face Representation Using Independent Component Analysis. Pattern Recognition, 2002, 35:1247-1257
[26] D.Q. ZHANG, S.C. CHEN, Z.H. ZHOU. A new face recognition method based on SVD perturbation for single example image per person. Applied Mathematics and Computation, 2005,163(2): 895-907
[27]庞彦伟,刘政凯,愈能海,等.融合奇异值分解和线性鉴别分析的人脸识别算法.电路与系统学报, 2006,11(4):47-50
[28] T. ALESANDER, Y. LI, L.J.ZHANG. Exact k-NN queries on clustered SVD datasets. Information Processing Letters, 2005,94(6):247-252
[29] J. FORTUNA, D. CAPSON. Improved support vector classification using PCA and ICA feature space modification. Pattern Recognition, 2007,37(6):1117-1129
[30] J. YANG, X.M. Gao, D. ZHANG, J.Y. YANG. Kernel ICA: An alternative formulation and its application to face recognition. Pattern Recognition, 2005,38(10):1784-1787
[31]蒋宗礼.人工神经网络导论.北京:高等教育出版社, 2002
[32] J. ZHANG, Y. YAN, M.LADES. Face reognition: Eigenfaces, Elastic Matching, and Neural Nets. IEEE Proc. 1998,85(9):1423-1435
[33] F.SAMARIA, S.YOUNG. HMM based architecture for face identification. Image and Computer Vision, 1994,12(4):537-543
[34] L.R. RABINER. A tutorial on hidden markov models and selected application in speech recognition. Proceedings of IEEE, 1989,77(2):257-285
[35] L.WISKOTT, J.M.FELLOUS, N. KRUEGER, etal. Face recognition by elastic bunch graph matching. IEEE Trans, On PAMI, 1997,63(3):665-779
[36] J.Z. MING, J.LIN, X.F. FAN. Feature separation using ICA for a one-dimensional time series and its application in fault detection. Journal of Sound and Vibration, 2005,3(287): 614-624
[37] D. RAMASUBRAMANIAN, Y.V. VENKATESH. Encoding and recognition of faces based on the human visual model and DCT. Pattern Recognition, 2001,34(12):2447-2458
[38]孙永宣,何柯峰,胡良梅.一种新的基于DCT变换的人脸特征.合肥工业大学学报(自然科学版), 2006,29(11):1396-1399
[39] K.ASSALEH, Y.ASSAF. Fearures extraction and analysis for classifying causable patterns in control charts. Computers & Industrial Engineering, 2005,49(1):168-181
[40]赵传强.基于DCT的人脸识别方法研究. [山东大学硕士学位论文]. 2007:40-41
[41]阮秋琦.数字图像处理学.北京:电子工业出版社, 2001.1
[42]杨淑莹. VC++图像处理程序设计.北京:清华大学出版社;北京交通大学出版社,2005.1
[43]周长发.精通Visual C++图像处理编程.北京:电子工业出版社,2000
[44]郎锐.数字图像处理学VC++实现.北京:北京希望电子出版社, 2002
[45]席庆,张春林. Visual C++ 6.0实用编程技术.北京:中国水利水电出版社, 1999
[46]张志军,孙志辉.基于VC平台的彩色图像的灰度化技术.自动化技术与应用.2005,24(1):61-63
[47]李文革.基于主成分分析的人脸识别. [山东大学硕士学位论文],2008: 23-25
[48]任艳斐.直方图均衡化在图像处理中的应用.科技信息, 2007(4):37-38
[49] H.WANG, R.A.HADDAD. Adaptive median filters: New algorithms and results. IEEE Transactions on Image Processing, 1995, 4(4):499-502
[50] I.DAUBESCHIES. Ten Lectures on Wavelets. Philadephia: SIAM Publ, 1993
[51]李世雄.小波变换及其应用.北京:高等教育出版社,1997
[52]马维祯,殷端祥.子波分析与子波变换.广州:华南理工大学出版社, 1996
[53] Z. M.HAFED, M. D.LEVIEN. Face recognition using the discrete cosine transform. Int. J. Comput. Vis, 2001,43(3):167-188
[54] Z.PAN, R.ADAMS, H.BOLOURI. Image redundancy reduction for neural network classification using discrete cosine transforms. In: Proceedings of the IEEE-INNS-ENNS International Joint Conf. On Neural Networks,Como, 2000,Italy 3:149-154.
[55] A.JAIN. Fundamentals of Digital Image Processing. Prentice: Englewood Cliffs,NJ. 1989.