基于类Harr特征和最小包含球的纸币识别方法的研究
|
摘要
目前在世界上每天流通的纸币数量巨大,在金融部门内部纸币整理工作是非常繁重的,如何快速准确的纸币清分在银行业中具有非常重要的意义。通过使用灵敏准确的纸币清分机,能使繁琐的钞票清分工作变得简易、快捷和可靠。然而国内清分机市场上出现的国产清分机,大都因为纸币识别率和拒识类的问题而影响了它们在国内的销售市场。如何在提高纸币图像的识别率的同时有效地降低拒识类样本被误识的几率是现代纸币图像处理中的难题,也是继续研究纸币清分技术时亟待解决的问题。
本文提出了基于类Harr特征和最小包含球的纸币图像识别方法。通过类Harr特征提取方法提取出多样化且更具灵活性的特征,并且通过特征选择降低识别特征的维数,保证识别特征的鉴别能力;另一方面,通过最小包含球方法解决其它各种纸币识别方法不能有效地拒识拒识类样本的问题。
类Harr特征法通过改进原始的两类类Harr特征提取方法、AdaBoost特征选择方法以及多币种识别方法以得到四种能用于多币种纸币识别的类Harr特征提取方法:整幅图像范围类Harr特征差值法、整幅图像范围类Harr特征求和法以及固定网格范围类Harr特征差值法和固定网格范围类Harr特征求和法。在实验中通过比较它们的效果,从中选择识别率最高的类Harr特征提取方法,改进了原始的特征提取方式。
最小包含球方法通过将同类样本限制在一个最优的最小包含球内而不同类别限制在不同的包含球内的方法,将拒识类的样本限制在任何有效类别的包含球之外,从而有效解决原先的识别方法不能有效地拒识拒识类样本的问题。经证明: SVM中的One Class方法在RBF核和等模长条件下的线性核与最小包含球是等价的。实验中用One Class方法来实现最小包含球方法。
实验表明:用类Harr特征提取法能提取出更富含“信息”的特征,在将这些特征用于识别时也能够有效地提高识别率;用最小包含球方法能有效地解决拒识类样本的识别问题。
Currently in the world, There are large numbers of paper currency circulates every day, it is a tough work for staves to sort the paper currency in the banking sector. How to sort the paper currency quickly and correctly becomes very important in the banks. It makes the cumbersome work simply, fast and reliably to use the sensitive and accurate paper currency sorter. The technology is to process and recognize national paper currency. However, the paper currency sorter yielded home in the home market has low sales ratio, mostly because of recognition rate、other issues such as low reject ratio of rejection notes. How to improve the image recognition rate and reduce the error recognition ratio of reject types is currently a puzzle, but also a urgent problem on the way of continuing to study the technology of paper currency processing.
In this paper, recognition methods of paper currency based on the Harr-like feature and the minimal ball including most of the same samples is proposed. The extraction of Harr-like feature extracted features different from the features extracted through other effectively feature extraction method, which is single and lack of flexibility, it can also reduce the dimension of feature and ensure the the ability of classification. On the other hand, the method of the minimal ball including most samples brings a new solution for the reject types which can’t be rejected by the other recognition methods.
The Harr-like feature extraction brings forward four recognition methods of paper currency which include Harr-like Minus method in the range of the whole image、Harr-like Sum method in the range of whole image、Harr-like Minus method in the range of fixed gridding and Harr-like Sum method in the range of fixed gridding through modifying the original two-types recognition method、the feature selection of AdaBoost and recognition methods for multiple types. A appropriate method which shows best performance in the test can be found.
The methods of minimal ball including most of the same samples while different types in different balls can exclude the reject samples which should be rejected, and so it can solve the problem that the original recognition methods can’t effectively reject the reject types. It’s proved that the method of One Class is equivalent the minimal ball including most samples, so we can use the method of One Class in SVM to realise it.
Experiments show that: the extraction of Harr-like feature can extract features which include more rich information, and it can also be used to identify paper currency ,which can effectively raise the recognition rate; the minimum ball including most samples is also a effective way, which can solve the reject problem.
本文提出了基于类Harr特征和最小包含球的纸币图像识别方法。通过类Harr特征提取方法提取出多样化且更具灵活性的特征,并且通过特征选择降低识别特征的维数,保证识别特征的鉴别能力;另一方面,通过最小包含球方法解决其它各种纸币识别方法不能有效地拒识拒识类样本的问题。
类Harr特征法通过改进原始的两类类Harr特征提取方法、AdaBoost特征选择方法以及多币种识别方法以得到四种能用于多币种纸币识别的类Harr特征提取方法:整幅图像范围类Harr特征差值法、整幅图像范围类Harr特征求和法以及固定网格范围类Harr特征差值法和固定网格范围类Harr特征求和法。在实验中通过比较它们的效果,从中选择识别率最高的类Harr特征提取方法,改进了原始的特征提取方式。
最小包含球方法通过将同类样本限制在一个最优的最小包含球内而不同类别限制在不同的包含球内的方法,将拒识类的样本限制在任何有效类别的包含球之外,从而有效解决原先的识别方法不能有效地拒识拒识类样本的问题。经证明: SVM中的One Class方法在RBF核和等模长条件下的线性核与最小包含球是等价的。实验中用One Class方法来实现最小包含球方法。
实验表明:用类Harr特征提取法能提取出更富含“信息”的特征,在将这些特征用于识别时也能够有效地提高识别率;用最小包含球方法能有效地解决拒识类样本的识别问题。
Currently in the world, There are large numbers of paper currency circulates every day, it is a tough work for staves to sort the paper currency in the banking sector. How to sort the paper currency quickly and correctly becomes very important in the banks. It makes the cumbersome work simply, fast and reliably to use the sensitive and accurate paper currency sorter. The technology is to process and recognize national paper currency. However, the paper currency sorter yielded home in the home market has low sales ratio, mostly because of recognition rate、other issues such as low reject ratio of rejection notes. How to improve the image recognition rate and reduce the error recognition ratio of reject types is currently a puzzle, but also a urgent problem on the way of continuing to study the technology of paper currency processing.
In this paper, recognition methods of paper currency based on the Harr-like feature and the minimal ball including most of the same samples is proposed. The extraction of Harr-like feature extracted features different from the features extracted through other effectively feature extraction method, which is single and lack of flexibility, it can also reduce the dimension of feature and ensure the the ability of classification. On the other hand, the method of the minimal ball including most samples brings a new solution for the reject types which can’t be rejected by the other recognition methods.
The Harr-like feature extraction brings forward four recognition methods of paper currency which include Harr-like Minus method in the range of the whole image、Harr-like Sum method in the range of whole image、Harr-like Minus method in the range of fixed gridding and Harr-like Sum method in the range of fixed gridding through modifying the original two-types recognition method、the feature selection of AdaBoost and recognition methods for multiple types. A appropriate method which shows best performance in the test can be found.
The methods of minimal ball including most of the same samples while different types in different balls can exclude the reject samples which should be rejected, and so it can solve the problem that the original recognition methods can’t effectively reject the reject types. It’s proved that the method of One Class is equivalent the minimal ball including most samples, so we can use the method of One Class in SVM to realise it.
Experiments show that: the extraction of Harr-like feature can extract features which include more rich information, and it can also be used to identify paper currency ,which can effectively raise the recognition rate; the minimum ball including most samples is also a effective way, which can solve the reject problem.
引文
1刘家锋,刘松波,唐降龙,一种实时纸币识别方法的研究,计算机研究与发展, 2003,40(7): 1000~1239
2张平,徐问之,基于神经网络多国货币种类的识别与研究,重庆大学学报(自然科学版),1999,22(3):1~10
3张国华,梁中华,一种基于模板匹配的人民币纸币面额识别方法,沈阳工业大学学报,2005,27(4): 439~442
4谢凯,郝建新,一种基于货币特征的神经网络模式识别算法,计算机应用,2002 , 22(11): 76~77
5尤佳,徐炜,流通人民币纸币的面值识别,仪器仪表学报,2003,24(4):1~5
6 Paul Viola and Michael Jones, Rapid Object Detection using a Boosted Cascade of Simple Features, International conference on computer vision and pattern recognition, Hawaii: IEEE, 2001:511~518
7 Chih-Chung Chang and Chih-Jen Lin_, LIBSVM: a Library for Support Vector Machines, April 17, 2005, Neural Compute,1998, (10):1299~1319
8 Bernhard Sch¨olkopf, Estimating the Support of a High-Dimensional Distribution, Microsoft Research Corporation Technical Report MSR-TR-99-87, 1999:20~100
9 John C. Platt and John Shawe-Taylor, Neural Computation , 2001,13: 1443~1471
10 Paul Viola, Michael Jones, Detecting Pedestrians Using Patterns of Motion and Appearance, Daniel Snow, The 9th Int' 1 Conf,Computer Vision, 2003, 1:734~741
11 Christopher J.C. Burges, A Tutorial on Support Vector Machines for Pattern Recognition, Data Mining and Knowledge Discovery,1998, 2:121~167
12 Keerthi S, Shevade S, Bhattcharyya C, Murthy K. Improvements to Platt's SMO algorithm for SVM classifier design. Neural Computation, 2001, 13(3):637~649
13 Vladimir Vapnik , Advances in Kernel Methods-Support Vector Learning—Preface, Kernel principal component analysis,Eds. Cambridge, MA: MIT Press, 1999:196~201
14 John C. Platt, Probabilistic Outputs for Support Vector Machines and Comparisons to Regularized Likelihood Methods, MA: MIT Press. 1999:2~30
15 Alex J. Smolay and Bernhard Sch¨olkopfz, A Tutorial on Support Vector Regression,Royal Holloway College, London,UK, 1998
16 Tao Qing, Sun Demin, Fan Jinsong et al. Maximal margin linear classifier based on the contraction of the closed convex hull. Journal of Software,2002, 13(3):404~409
17 John C. Platt,Fast Training of Support Vector Machines using Sequential Minimal Optimization,Advances in Kernel Methods - Support Vector Learning. Cambridge , MA: MIT Press,1999:185~208
18 David M.J. Tax *,1, Robert P.W. Duin,Sequential Minimal Optimization for SVM Support vector domain description,www.ics.uci.edu/xge/svm/smo.html,2001
19李萍萍.周贵祥,基于货币特征的模式识别方法研究,电气技术与自动化,机械制造与自动化2006,(2):204~302
20 John Shawe-Taylor Nello Christian,模式分析的核方法,赵玲玲,翁苏明,曾华军等译,机械工业出版社,2006:126~163
21 F Takeda, S Omatu.,“Bank Note Recognition System Using Neural Network with Random Masks,”Proceeding of the World Congress an Neural Networks, Portland, USA, 1993, 1:241~244
22何佳兵,李习伦,刘松波,王希奎.纸币清分机系统的研究与开发,机电工程技术,2007, 36 (8):29~32
23 F Takeda, S Omatu,“Neural Network Systems Technique and Applications in Paper Currency Recognition”, Neural Network Systems, Techniques and Applications, ACADEMIC Press, 1998, 5:133~160
24 A. Frosini, M. Gori, and P. Priami,“A neural network based model for paper currency recognition and verification,”IEEE Trans. Neural Networks, 1996 7:1482~1490
25 A. Ahmadi and S. Omatu,“A High Reliabilily Method for Classification of Paper Currency Based on Neural Networks,”Proceeding of The Eighth International Symposium an Artificial Life and Robotics (AROB 8th‘03) Oita, Japan, 2003, 2:601~604
26 F Takeda, S Omatu. High speed paper currency recognition by neural networks. IEEE Trans on Neural Network, 1995, 6(1):73~77
27 F Takeda, S Omatu,“A neuro-money recognition using optimized masks by CA”,Advance in Fuzzy Logic, Neural Networks and Genetic Algorithms LNAI 1011, Springer,1995:190~201
28崔明星,纸币识别算法的研究,哈尔滨工业大学硕士学位论文2002
29孔繁辉,高速清分机人民币面值面向识别技术的研究,哈尔滨工业大学硕士学位论文2004
30 T. Cootes, C. Taylor, D. Cooper, and J. Graham,“Active Shape Image Understanding, Models—Their Training and Application,”Computer Vision, 1995, 61:38~59
31 P. Moerland. A Comparison of Mixture Models for Density Estimation Artifical Neural Networks. Conference Publication. 1999,(470):7~10
32 F.Takeda, Sigeru Omatu. High Speed Paper Currency Recognition by Neural Networks. IEEE Trans Neural Network. 1995, l6(1)
33 Frosini, M. Gori, P. Priami. A Neural Network-Based Model for Paper Currency Recognition and Verification. IEEE Transactions on Neural Networks. 1996, 7(6):1482~1490
34 F. Takeda, S. Omatu, T. Inoue, and S. Onami. A Structure Reduction of Neural Network with Random Masks and Bill Money Recognition. Proceedings of the 2nd International Conference on Fuzzy Logic and Neural Networks. 1999, 12(5)
35 Andrew Webb. Statistical Pattern Recognition. Oxford University Press, New York, 1999
36 C. Papageorgiou, M. Oren, and T. Poggio. A general framework for object detection. In International Conference on Computer Vision, 1998
37 A D. Roth, M. Yang, and N. Ahuja. A snowbased face detector. In Neural Information Processing, 2000,12
38 Yoav Freund and Robert E. Schapire. A decision-theoretic generalization of on-line learning and an application to boosting. In Computational Learning Theory, Eurocolt,. Springer-Verlag, 1995:23~37
39 F Takeda, S Omatu. A neuro-system technology for bank note recognition, Proceedings of the JapanNSA Symposium on Flexible Automation, Boston, USA, 1996, 2: 1511~1516
40 A. Ahmadi and S. Omatu,“A High Reliabilily Method for Classification of Paper Currency Based on Neural Networks,”Proceeding of The Eighth International Symposium an Artificial Life and Robotics, Oita, Japan, 2003, 2:601~604
41 M. Tanaka,“Modeling of Mixtures of Principal Component Analysis Modcl with Genetic Algorithm,”Proc. 3lst Int’l Symposium on Stochastic Systems Thcory and Its Applications, Yokohama, Japan, 1999:157-162
42 M. Teranishi, S. Omahl and T. Kosaka,“Neura-classifier ofcurrency Fatigue Level Based on Acoustic Cepstlum Pattems,”Joumal of Advanced Computational Intelligence, 2000,4(1):18-23
2张平,徐问之,基于神经网络多国货币种类的识别与研究,重庆大学学报(自然科学版),1999,22(3):1~10
3张国华,梁中华,一种基于模板匹配的人民币纸币面额识别方法,沈阳工业大学学报,2005,27(4): 439~442
4谢凯,郝建新,一种基于货币特征的神经网络模式识别算法,计算机应用,2002 , 22(11): 76~77
5尤佳,徐炜,流通人民币纸币的面值识别,仪器仪表学报,2003,24(4):1~5
6 Paul Viola and Michael Jones, Rapid Object Detection using a Boosted Cascade of Simple Features, International conference on computer vision and pattern recognition, Hawaii: IEEE, 2001:511~518
7 Chih-Chung Chang and Chih-Jen Lin_, LIBSVM: a Library for Support Vector Machines, April 17, 2005, Neural Compute,1998, (10):1299~1319
8 Bernhard Sch¨olkopf, Estimating the Support of a High-Dimensional Distribution, Microsoft Research Corporation Technical Report MSR-TR-99-87, 1999:20~100
9 John C. Platt and John Shawe-Taylor, Neural Computation , 2001,13: 1443~1471
10 Paul Viola, Michael Jones, Detecting Pedestrians Using Patterns of Motion and Appearance, Daniel Snow, The 9th Int' 1 Conf,Computer Vision, 2003, 1:734~741
11 Christopher J.C. Burges, A Tutorial on Support Vector Machines for Pattern Recognition, Data Mining and Knowledge Discovery,1998, 2:121~167
12 Keerthi S, Shevade S, Bhattcharyya C, Murthy K. Improvements to Platt's SMO algorithm for SVM classifier design. Neural Computation, 2001, 13(3):637~649
13 Vladimir Vapnik , Advances in Kernel Methods-Support Vector Learning—Preface, Kernel principal component analysis,Eds. Cambridge, MA: MIT Press, 1999:196~201
14 John C. Platt, Probabilistic Outputs for Support Vector Machines and Comparisons to Regularized Likelihood Methods, MA: MIT Press. 1999:2~30
15 Alex J. Smolay and Bernhard Sch¨olkopfz, A Tutorial on Support Vector Regression,Royal Holloway College, London,UK, 1998
16 Tao Qing, Sun Demin, Fan Jinsong et al. Maximal margin linear classifier based on the contraction of the closed convex hull. Journal of Software,2002, 13(3):404~409
17 John C. Platt,Fast Training of Support Vector Machines using Sequential Minimal Optimization,Advances in Kernel Methods - Support Vector Learning. Cambridge , MA: MIT Press,1999:185~208
18 David M.J. Tax *,1, Robert P.W. Duin,Sequential Minimal Optimization for SVM Support vector domain description,www.ics.uci.edu/xge/svm/smo.html,2001
19李萍萍.周贵祥,基于货币特征的模式识别方法研究,电气技术与自动化,机械制造与自动化2006,(2):204~302
20 John Shawe-Taylor Nello Christian,模式分析的核方法,赵玲玲,翁苏明,曾华军等译,机械工业出版社,2006:126~163
21 F Takeda, S Omatu.,“Bank Note Recognition System Using Neural Network with Random Masks,”Proceeding of the World Congress an Neural Networks, Portland, USA, 1993, 1:241~244
22何佳兵,李习伦,刘松波,王希奎.纸币清分机系统的研究与开发,机电工程技术,2007, 36 (8):29~32
23 F Takeda, S Omatu,“Neural Network Systems Technique and Applications in Paper Currency Recognition”, Neural Network Systems, Techniques and Applications, ACADEMIC Press, 1998, 5:133~160
24 A. Frosini, M. Gori, and P. Priami,“A neural network based model for paper currency recognition and verification,”IEEE Trans. Neural Networks, 1996 7:1482~1490
25 A. Ahmadi and S. Omatu,“A High Reliabilily Method for Classification of Paper Currency Based on Neural Networks,”Proceeding of The Eighth International Symposium an Artificial Life and Robotics (AROB 8th‘03) Oita, Japan, 2003, 2:601~604
26 F Takeda, S Omatu. High speed paper currency recognition by neural networks. IEEE Trans on Neural Network, 1995, 6(1):73~77
27 F Takeda, S Omatu,“A neuro-money recognition using optimized masks by CA”,Advance in Fuzzy Logic, Neural Networks and Genetic Algorithms LNAI 1011, Springer,1995:190~201
28崔明星,纸币识别算法的研究,哈尔滨工业大学硕士学位论文2002
29孔繁辉,高速清分机人民币面值面向识别技术的研究,哈尔滨工业大学硕士学位论文2004
30 T. Cootes, C. Taylor, D. Cooper, and J. Graham,“Active Shape Image Understanding, Models—Their Training and Application,”Computer Vision, 1995, 61:38~59
31 P. Moerland. A Comparison of Mixture Models for Density Estimation Artifical Neural Networks. Conference Publication. 1999,(470):7~10
32 F.Takeda, Sigeru Omatu. High Speed Paper Currency Recognition by Neural Networks. IEEE Trans Neural Network. 1995, l6(1)
33 Frosini, M. Gori, P. Priami. A Neural Network-Based Model for Paper Currency Recognition and Verification. IEEE Transactions on Neural Networks. 1996, 7(6):1482~1490
34 F. Takeda, S. Omatu, T. Inoue, and S. Onami. A Structure Reduction of Neural Network with Random Masks and Bill Money Recognition. Proceedings of the 2nd International Conference on Fuzzy Logic and Neural Networks. 1999, 12(5)
35 Andrew Webb. Statistical Pattern Recognition. Oxford University Press, New York, 1999
36 C. Papageorgiou, M. Oren, and T. Poggio. A general framework for object detection. In International Conference on Computer Vision, 1998
37 A D. Roth, M. Yang, and N. Ahuja. A snowbased face detector. In Neural Information Processing, 2000,12
38 Yoav Freund and Robert E. Schapire. A decision-theoretic generalization of on-line learning and an application to boosting. In Computational Learning Theory, Eurocolt,. Springer-Verlag, 1995:23~37
39 F Takeda, S Omatu. A neuro-system technology for bank note recognition, Proceedings of the JapanNSA Symposium on Flexible Automation, Boston, USA, 1996, 2: 1511~1516
40 A. Ahmadi and S. Omatu,“A High Reliabilily Method for Classification of Paper Currency Based on Neural Networks,”Proceeding of The Eighth International Symposium an Artificial Life and Robotics, Oita, Japan, 2003, 2:601~604
41 M. Tanaka,“Modeling of Mixtures of Principal Component Analysis Modcl with Genetic Algorithm,”Proc. 3lst Int’l Symposium on Stochastic Systems Thcory and Its Applications, Yokohama, Japan, 1999:157-162
42 M. Teranishi, S. Omahl and T. Kosaka,“Neura-classifier ofcurrency Fatigue Level Based on Acoustic Cepstlum Pattems,”Joumal of Advanced Computational Intelligence, 2000,4(1):18-23
© 2004-2018 中国地质图书馆版权所有 京ICP备05064691号 京公网安备11010802017129号 地址：北京市海淀区学院路29号 邮编：100083 电话：办公室：(+86 10)66554848；文献借阅、咨询服务、科技查新：66554700 |