基于SVM的车牌字符识别研究
摘要
在社会经济高速发展的今天,各种车辆的数目随之急剧增加。智能交通系统(ITS)已经成为各国用来解决日益突出的公路交通问题、进行城市交通管理的主要手段。而车牌照自动识别(LPR)则是智能交通系统中极为重要的一部分。其中,字符的正确识别率和识别速度则是衡量车牌识别系统性能的主要指标。
鉴于车牌字符识别的技术的重要性,许多学者都已投身该领域,并做了一些富有成效的工作。但许多传统的基于图像处理的字符识别技术,如模板匹配方法,人工神经网络方法等,都表现出一些局限性:需要大量的字符样本,推广能力较差,正确识别率偏低,难以真正投入到实际工程应用中去。
统计学习理论(SLT)是建立在小样本学习基础上的,它为研究有限样本情况下的统计模式识别建立了一个较好的理论框架,并推出了一种新的模式识别方法──支持向量机(SVM)。SVM能较好地克服诸如局部最优解、维数灾难等困扰传统模式识别方法的难题,且在小样本条件下体现出很好的推广性能。
本文着重研究了支持向量机在车牌字符识别中的应用。首先介绍了统计学习理论和支持向量机的一些基本知识,较深入的探讨了支持向量机的学习算法,然后研究了字符特征的表示方法,提出了字符的小波网格特征和投影特征,最后编程完成了一个基于SVMLight算法的车牌字符识别软件。该软件在小样本情况下获得了95%以上的正确识别率。
Nowadays, with the high-speed development of socio-economic, the amount of vehicles has increased dramatically. The Intelligent Transport System(ITS)has become the main instrument of solving the problems of highway traffic and the traffic control of cities. And the License Plate Recognition (LPR) plays a main role in the ITS, in which the accuracy and speed of character recognition are key indicators.
In view of the great significance of LPR technology, many researchers turn to work on it and achieved some positive results. However, Many of the traditional ideas, which are based on the image processing technology, such as template matching method and artificial neural network method, have shown some limitations: the need of large amount of samples, poor performance of promotion capacity, and low accuracy of character recognition. It is difficult to put them into practical applications.
The Statistical Learning Theory (SLT) is built on the case of small samples, it supplies a better theoretical frame to the research of Statistical Pattern Recognition under the circumstances of limited samples, and presents a new pattern recognition method──the Support Vector Machine (SVM).The SVM can overcome problems such as the local optimal solution, dimension disaster and other difficulties which trouble the traditional pattern recognition methods badly. It also shows good promotion capacity in the case of small samples.
This thesis focuses on the application of Support Vector Machine in character recognition of license plate. Firstly, some basic knowledge of the Statistical Learning Theory and Support Vector Machine is introduced. The training algorithms of Support Vector Machine are also discussed in detail. Then, the methods of character feature are researched; the wavelet grid feature and projection feature of character are raised. Finally, a character recognition software based on SVMLight algorithm is programmed. This software gains over 95% accuracy while the number of samples is rather small.
鉴于车牌字符识别的技术的重要性,许多学者都已投身该领域,并做了一些富有成效的工作。但许多传统的基于图像处理的字符识别技术,如模板匹配方法,人工神经网络方法等,都表现出一些局限性:需要大量的字符样本,推广能力较差,正确识别率偏低,难以真正投入到实际工程应用中去。
统计学习理论(SLT)是建立在小样本学习基础上的,它为研究有限样本情况下的统计模式识别建立了一个较好的理论框架,并推出了一种新的模式识别方法──支持向量机(SVM)。SVM能较好地克服诸如局部最优解、维数灾难等困扰传统模式识别方法的难题,且在小样本条件下体现出很好的推广性能。
本文着重研究了支持向量机在车牌字符识别中的应用。首先介绍了统计学习理论和支持向量机的一些基本知识,较深入的探讨了支持向量机的学习算法,然后研究了字符特征的表示方法,提出了字符的小波网格特征和投影特征,最后编程完成了一个基于SVMLight算法的车牌字符识别软件。该软件在小样本情况下获得了95%以上的正确识别率。
Nowadays, with the high-speed development of socio-economic, the amount of vehicles has increased dramatically. The Intelligent Transport System(ITS)has become the main instrument of solving the problems of highway traffic and the traffic control of cities. And the License Plate Recognition (LPR) plays a main role in the ITS, in which the accuracy and speed of character recognition are key indicators.
In view of the great significance of LPR technology, many researchers turn to work on it and achieved some positive results. However, Many of the traditional ideas, which are based on the image processing technology, such as template matching method and artificial neural network method, have shown some limitations: the need of large amount of samples, poor performance of promotion capacity, and low accuracy of character recognition. It is difficult to put them into practical applications.
The Statistical Learning Theory (SLT) is built on the case of small samples, it supplies a better theoretical frame to the research of Statistical Pattern Recognition under the circumstances of limited samples, and presents a new pattern recognition method──the Support Vector Machine (SVM).The SVM can overcome problems such as the local optimal solution, dimension disaster and other difficulties which trouble the traditional pattern recognition methods badly. It also shows good promotion capacity in the case of small samples.
This thesis focuses on the application of Support Vector Machine in character recognition of license plate. Firstly, some basic knowledge of the Statistical Learning Theory and Support Vector Machine is introduced. The training algorithms of Support Vector Machine are also discussed in detail. Then, the methods of character feature are researched; the wavelet grid feature and projection feature of character are raised. Finally, a character recognition software based on SVMLight algorithm is programmed. This software gains over 95% accuracy while the number of samples is rather small.
引文
[1]黄卫,陈里得.智能运输系统(ITS)概述.北京:人民交通出版社,2001.
[2]马俊莉,莫玉龙,王明祥.一种基于改进模板匹配的车牌字符识别方法.小型微型计算机系统, 2003,24(9): 1670-1672.
[3]王敏,黄心汉,魏武等.一种模板匹配和神经网络的车牌字符识别办法.华中科技大学学报, 2001,29(3).
[4]王春,刘波,周新志.采用BP神经网络的车牌字符识别方法研究.中国测试技术,2005,31(1): 26-28.
[5]关山,王伟杰,赵学增.基于BP神经网络的车牌字符自动识别方法的研究.东北电力学院学报, 2004,24(4): 56-60.
[6]李波.基于数学形态学和神经网络的车牌字符识别系统研究:[硕士学位论文]华中科技大学,2006.
[7]胡爱民.复杂情况下的车牌自动识别技术研究:[博士学位论文]北京航空航天大学,2001.
[8] V. Vapnik. The Nature of Statistical Learning Theory. Springer-Verlag, New York. 1995.
[9] M. O. Stitson, J. A. E. Weston, A. Gammerman, V. Vovk and V. Vapnik. Theory of Support Vector Machines. Technical Report CSD-TR-96-17, Computational Intelligence Group, Royal Holloway, University of London, 1996.
[10]边肇祺,张学工.模式识别(第二版).北京:清华大学出版社,2000.
[11] T. Joachims. Transductive Inference for Text Classification using Support Vector Machines. International Conference on Machine Learning (ICML), 1999.
[12]宁朝.基于SVM技术的英文字符识别方法.大众科技,2005,8: 160-161.
[13]石繁槐,童学锋.SVM在小字符集脱机手写体汉字识别中的应用研究.计算机工程,2002,28(6): 154-155,189.
[14] Yongjun Ma, Tingjian Fang, Kai Fang, Dingcheng Wang, Wei Chen. Texture imageclassification based on support vector machine and distance classification. Proceedings of the 4th World Congress on Intelligent Control and Automation, 2002.
[15] C.J.C.Burges. A Tutorial on Support Vector Machine for Pattern Rcgonition, Knowledge Discovery and Data Mining,Vol.2,1998.
[16] K. Nigam, A. McCallum, and T. Mitchell. Learning to classify text from labeled and unlabeled documents. In Proceedings of AAAI-1998.
[17]黄志斌,陈锻生.支持向量机在车牌字符识别中的应用.计算机工程,2003,29(5): 192-194.
[18]任俊.基于支撑矢量机的图像分类、车牌识别及嵌入式应用研究:[博士学位论文]浙江大学,2005年.
[19]田盛丰.基于核函数的学习算法.北方交通大学学报,2003,27(2): 1-8.
[20]王鹏,朱小燕.基于RBF核的SVM的模型选择及其应用.清华大学计算机系智能技术与系统国家重点实验室,2003.
[21] J. Platt. Sequential minimal optimization: A fast algorithm for training support vector machines. Technical Report MSR-TR-98-14, Microsoft Research, 1998.
[22]兰光华.支持向量机训练算法实现及其改进:[硕士学位论文]南京理工大学,2005.
[23] Kuanming Lin, Chihjen Lin. A study on reduced support vector machines. IEEE Transactions on NEURAL Networks,14(6), 2003.
[24] E.Osuna, R.Ferund, and F.Girosi. Support vector machines: Training and applications. AI Memo 1602, Massachusetts Insittute of Technology, 1997a.
[25] D.Lai,N.Mani,and M.Palaniswami. A new method to select working sets for faster training for support vector machines.Technical Report MESCE-30-2003, Dept.Electrical and Computer Systems Engineering Monash University,Australia,2003b.
[26]刘华富.一种快速支持向量机分类算法.长沙大学学报, 2004,18(4): 40-41.
[27] C.J.Lin. Asymptotic convergence of an SMO algorithm without any assumptions.IEEE Transactions on Neural Networks,13(1):248-250,2002.
[28] Jianmin Li, Bo Zhang, Fuzong Lin. A New Cache Replacement Algorithmin SMO.Lecture Notes In Computer Science, 2388:342-353.
[29] C.C.Chang and C.J.Lin. LIBSVM:a library for support vector machines,2005.
[30] T. Joachims. SVMlight--Support Vector Machine. http://www-ai.cs.uni- dortmund.de/ SOFTWARE/SVM_LIGHT/svm_light.eng.html.
[31] Laura Palagi and Marco Sciandrone. On the convergence of a modified version of SVMlight algorithm.Opitmization Methods and Software, 20(2-3):315-332, 2005.
[32]陈光英,张千里,李星.特征选择和SVM训练模型的联合优化,清华大学学报(自然科学版),2004,44(1): 9-12.
[33]柳回春,马树元,吴平东等.手写体数字识别技术的研究.计算机工程, 2003,29(4): 24-25,61.
[34]靳济芳.Visual C++小波变换技术与工程实践.北京:人民邮电出版社,2004.
[35]唐圣学,何仪刚.基于小波预处理的神经网络字符识别方法.吉首大学学报(自然科学版),2005,26(2): 14-17.
[36]王志红.小波和神经网络模式识别技术及其在车牌识别中的应用:[硕士学位论文]合肥工业大学,2003年.
[37]盛军.基于小波的图像分析及其在车牌识别中的应用:[硕士学位论文]合肥工业大学,2002.
[38]薛宏全.基于小波分析的车牌识别系统研究:[硕士学位论文]电子科技大学,2004.
[39]高珊,刘万春,朱玉文.基于SVM的车牌字符分割和识别方法.微电子学与计算机,2005,22(6): 34-36.
[40]任俊,李志能.支持向量机在字符分类识别中的应用.浙江大学学报(工学版), 2005,39(8): 1136-1141.
[41] Jiuzhen Liang. SVM multi-classifier and Web document classification. Proceedings of the 3rd International Conference on Machine Learning and Cybernetics,2004.
[42] Hengnian Qi, Jiangang Yang, Yiwen Zhong, Chao Deng. Multi-class SVM based remote sensing image classification and its semi-supervised improvement scheme. Proceedings of the 3rd International Conference on Machine Learning and Cybernetics,2004.
[43]王晓光,王晓华.一种基于SVM的车牌汉字的有效识别方法.计算机工程与应用,2004,40(24): 208-209,222.
[44]梅胜.基于SVM的多层次大类别数文本分类系统(HJ-TCM)的设计与实现: [硕士学位论文]南京理工大学,2004.
[45]陈毅松,汪国平,董士海.基于支持向量机的渐进直推式分类学习.北京大学计算机系人机交互与多媒体实验室.
[2]马俊莉,莫玉龙,王明祥.一种基于改进模板匹配的车牌字符识别方法.小型微型计算机系统, 2003,24(9): 1670-1672.
[3]王敏,黄心汉,魏武等.一种模板匹配和神经网络的车牌字符识别办法.华中科技大学学报, 2001,29(3).
[4]王春,刘波,周新志.采用BP神经网络的车牌字符识别方法研究.中国测试技术,2005,31(1): 26-28.
[5]关山,王伟杰,赵学增.基于BP神经网络的车牌字符自动识别方法的研究.东北电力学院学报, 2004,24(4): 56-60.
[6]李波.基于数学形态学和神经网络的车牌字符识别系统研究:[硕士学位论文]华中科技大学,2006.
[7]胡爱民.复杂情况下的车牌自动识别技术研究:[博士学位论文]北京航空航天大学,2001.
[8] V. Vapnik. The Nature of Statistical Learning Theory. Springer-Verlag, New York. 1995.
[9] M. O. Stitson, J. A. E. Weston, A. Gammerman, V. Vovk and V. Vapnik. Theory of Support Vector Machines. Technical Report CSD-TR-96-17, Computational Intelligence Group, Royal Holloway, University of London, 1996.
[10]边肇祺,张学工.模式识别(第二版).北京:清华大学出版社,2000.
[11] T. Joachims. Transductive Inference for Text Classification using Support Vector Machines. International Conference on Machine Learning (ICML), 1999.
[12]宁朝.基于SVM技术的英文字符识别方法.大众科技,2005,8: 160-161.
[13]石繁槐,童学锋.SVM在小字符集脱机手写体汉字识别中的应用研究.计算机工程,2002,28(6): 154-155,189.
[14] Yongjun Ma, Tingjian Fang, Kai Fang, Dingcheng Wang, Wei Chen. Texture imageclassification based on support vector machine and distance classification. Proceedings of the 4th World Congress on Intelligent Control and Automation, 2002.
[15] C.J.C.Burges. A Tutorial on Support Vector Machine for Pattern Rcgonition, Knowledge Discovery and Data Mining,Vol.2,1998.
[16] K. Nigam, A. McCallum, and T. Mitchell. Learning to classify text from labeled and unlabeled documents. In Proceedings of AAAI-1998.
[17]黄志斌,陈锻生.支持向量机在车牌字符识别中的应用.计算机工程,2003,29(5): 192-194.
[18]任俊.基于支撑矢量机的图像分类、车牌识别及嵌入式应用研究:[博士学位论文]浙江大学,2005年.
[19]田盛丰.基于核函数的学习算法.北方交通大学学报,2003,27(2): 1-8.
[20]王鹏,朱小燕.基于RBF核的SVM的模型选择及其应用.清华大学计算机系智能技术与系统国家重点实验室,2003.
[21] J. Platt. Sequential minimal optimization: A fast algorithm for training support vector machines. Technical Report MSR-TR-98-14, Microsoft Research, 1998.
[22]兰光华.支持向量机训练算法实现及其改进:[硕士学位论文]南京理工大学,2005.
[23] Kuanming Lin, Chihjen Lin. A study on reduced support vector machines. IEEE Transactions on NEURAL Networks,14(6), 2003.
[24] E.Osuna, R.Ferund, and F.Girosi. Support vector machines: Training and applications. AI Memo 1602, Massachusetts Insittute of Technology, 1997a.
[25] D.Lai,N.Mani,and M.Palaniswami. A new method to select working sets for faster training for support vector machines.Technical Report MESCE-30-2003, Dept.Electrical and Computer Systems Engineering Monash University,Australia,2003b.
[26]刘华富.一种快速支持向量机分类算法.长沙大学学报, 2004,18(4): 40-41.
[27] C.J.Lin. Asymptotic convergence of an SMO algorithm without any assumptions.IEEE Transactions on Neural Networks,13(1):248-250,2002.
[28] Jianmin Li, Bo Zhang, Fuzong Lin. A New Cache Replacement Algorithmin SMO.Lecture Notes In Computer Science, 2388:342-353.
[29] C.C.Chang and C.J.Lin. LIBSVM:a library for support vector machines,2005.
[30] T. Joachims. SVMlight--Support Vector Machine. http://www-ai.cs.uni- dortmund.de/ SOFTWARE/SVM_LIGHT/svm_light.eng.html.
[31] Laura Palagi and Marco Sciandrone. On the convergence of a modified version of SVMlight algorithm.Opitmization Methods and Software, 20(2-3):315-332, 2005.
[32]陈光英,张千里,李星.特征选择和SVM训练模型的联合优化,清华大学学报(自然科学版),2004,44(1): 9-12.
[33]柳回春,马树元,吴平东等.手写体数字识别技术的研究.计算机工程, 2003,29(4): 24-25,61.
[34]靳济芳.Visual C++小波变换技术与工程实践.北京:人民邮电出版社,2004.
[35]唐圣学,何仪刚.基于小波预处理的神经网络字符识别方法.吉首大学学报(自然科学版),2005,26(2): 14-17.
[36]王志红.小波和神经网络模式识别技术及其在车牌识别中的应用:[硕士学位论文]合肥工业大学,2003年.
[37]盛军.基于小波的图像分析及其在车牌识别中的应用:[硕士学位论文]合肥工业大学,2002.
[38]薛宏全.基于小波分析的车牌识别系统研究:[硕士学位论文]电子科技大学,2004.
[39]高珊,刘万春,朱玉文.基于SVM的车牌字符分割和识别方法.微电子学与计算机,2005,22(6): 34-36.
[40]任俊,李志能.支持向量机在字符分类识别中的应用.浙江大学学报(工学版), 2005,39(8): 1136-1141.
[41] Jiuzhen Liang. SVM multi-classifier and Web document classification. Proceedings of the 3rd International Conference on Machine Learning and Cybernetics,2004.
[42] Hengnian Qi, Jiangang Yang, Yiwen Zhong, Chao Deng. Multi-class SVM based remote sensing image classification and its semi-supervised improvement scheme. Proceedings of the 3rd International Conference on Machine Learning and Cybernetics,2004.
[43]王晓光,王晓华.一种基于SVM的车牌汉字的有效识别方法.计算机工程与应用,2004,40(24): 208-209,222.
[44]梅胜.基于SVM的多层次大类别数文本分类系统(HJ-TCM)的设计与实现: [硕士学位论文]南京理工大学,2004.
[45]陈毅松,汪国平,董士海.基于支持向量机的渐进直推式分类学习.北京大学计算机系人机交互与多媒体实验室.