基于Haar提升小波和SVM的离线笔迹鉴别
|
摘要
当今社会，生物识别技术的迅速发展，带动了手写体笔迹鉴别(Handwritingidentification,HWI)的发展，如今手写体笔迹鉴别已经成为计算机视觉和模式识别领域中的一个研究热点,而且基于笔迹的身份鉴别更是被广泛的应用在金融、社会化考试、银行甚至考古学等领域。
笔迹鉴别的本质就是根据手写笔迹来判断书写者的一门科学和技术。本文首先介绍了国内外笔迹鉴别的研究现状和相关理论，然后对现有算法进行了简要的介绍和分析,最后提出了基于Haar提升小波和支持向量机（SVM）的离线笔迹鉴别算法。该算法主要包括以下三个部分：
1.预处理。本文的预处理算法主要包含以下几个步骤：图像的灰度化，黑白二值化，去除噪声，单字切割，归一化处理和纹理形成等。
2.特征提取。本文的特征提取是基于纹理图进行的，分为两个部分,即：全局纹理特征提取和单字纹理特征提取。全局纹理特征提取，采用了基于二维Gabor变换的算法该算法用32个核函数（即4个频率和8个相位）进行仿真训练，通过与纹理图进行卷积运算，得到相应的32个小波变换系数，求其方差，将方差值作为全局纹理特征；单字纹理特征提取，采用了基于Haar提升小波变换的算法，该算法作为第二代小波变换，实现了从整数到整数的离散小波变换(DWT)，通过对单字样本进行三级小波分解，并求其小波系数的方差，得到单字特征。最后综合分析，给出最终鉴别结果。
3.分类器设计。本文采用SVM（支持向量机）进行分类，该分类器包括SVM的训练和SVM的分类两部分。在训练阶段，输入样本进行训练，保存训练结果；在分类阶段，输入测试样本和指定训练结果实现测试样本的分类。
本文选择40个人(每人2份共80份)的手写笔迹进行实验，以MATLAB7.0为实验环境，利用二维Gabor变换和Haar提升小波变换提取笔迹图像的纹理特征，再通过SVM分类器进行分类，完成笔迹鉴别的整个实验过程，并且取得了较好的实验结果。
Nowadays, Handwriting Identification (HWI), promoted by the rapid development of biologicalidentification technology, has become a research hotspot in computer Vision and Pattern Recognition. Andthe technologies based on HWI research develop well and have a wide range of applications in financialsystem, various social examinations, business bank and many other relevant areas.
Handwriting identification is such a technique that aims to identify one’s writing based on his or herhandwriting features. At first, this dissertation introduces the present research conditions of handwritingidentification and related theories, gives a survey of handwriting algorithm at home and abroad, and thenthis paper proposes the off-line HWI algorithm based on Haar lifting wavelet and SVM. The algorithmmainly includes three parts:
1. Pre-processing. Pre-processing in this paper mainly includes the following procedures: image grayprocessing, image binarization, noise removing, cutting words, normalization, texture formation and soon．
2. Feature extraction. Feature extraction in this paper is carried out based on texture map, and it isdivided into two parts: global texture feature extraction and single character texture feature extraction. Forglobal texture feature extraction, this paper proposes an algorithm based on two-dimensional Gabor filter.This algorithm uses32Kernel functions (4frequencies and8directions) for simulation training.32wavelettransform coefficients are gained through the convolution of texture, and then seek the variance value asglobal texture features. For single character texture feature extraction, this paper proposes the algorithmbased on Haar lifting wavelet. This algorithm, as second-generation wavelet transform, achieves DWT(discrete wavelet transforms) from integer to integer. Seek the variance of the wavelet coefficients throughdecomposing the single character sample into three levels and get the single character features. Aftercomprehensive analysis, the final results of the identification will be given.
3. Classifier design. This paper adopts the SVM classification algorithm to classify, including: SVMtraining and SVM classification. In the training phase, the input sample will be trained and the trainingresults will be saved; in the classification phase, inputting the test samples and specifying the trainingresults in order to achieve the classification of the test samples.
The experiment, selecting the handwriting samples of40individuals, is carried out in MATLAB7.0environment. Extract the texture features of the handwriting image by using2-dimensional Gabortransform and Haar lifting wavelet transform, and then complete the whole process of handwritingidentification after classifying by using the SVM classifier. The experiment achieves satisfactory results.
笔迹鉴别的本质就是根据手写笔迹来判断书写者的一门科学和技术。本文首先介绍了国内外笔迹鉴别的研究现状和相关理论，然后对现有算法进行了简要的介绍和分析,最后提出了基于Haar提升小波和支持向量机（SVM）的离线笔迹鉴别算法。该算法主要包括以下三个部分：
1.预处理。本文的预处理算法主要包含以下几个步骤：图像的灰度化，黑白二值化，去除噪声，单字切割，归一化处理和纹理形成等。
2.特征提取。本文的特征提取是基于纹理图进行的，分为两个部分,即：全局纹理特征提取和单字纹理特征提取。全局纹理特征提取，采用了基于二维Gabor变换的算法该算法用32个核函数（即4个频率和8个相位）进行仿真训练，通过与纹理图进行卷积运算，得到相应的32个小波变换系数，求其方差，将方差值作为全局纹理特征；单字纹理特征提取，采用了基于Haar提升小波变换的算法，该算法作为第二代小波变换，实现了从整数到整数的离散小波变换(DWT)，通过对单字样本进行三级小波分解，并求其小波系数的方差，得到单字特征。最后综合分析，给出最终鉴别结果。
3.分类器设计。本文采用SVM（支持向量机）进行分类，该分类器包括SVM的训练和SVM的分类两部分。在训练阶段，输入样本进行训练，保存训练结果；在分类阶段，输入测试样本和指定训练结果实现测试样本的分类。
本文选择40个人(每人2份共80份)的手写笔迹进行实验，以MATLAB7.0为实验环境，利用二维Gabor变换和Haar提升小波变换提取笔迹图像的纹理特征，再通过SVM分类器进行分类，完成笔迹鉴别的整个实验过程，并且取得了较好的实验结果。
Nowadays, Handwriting Identification (HWI), promoted by the rapid development of biologicalidentification technology, has become a research hotspot in computer Vision and Pattern Recognition. Andthe technologies based on HWI research develop well and have a wide range of applications in financialsystem, various social examinations, business bank and many other relevant areas.
Handwriting identification is such a technique that aims to identify one’s writing based on his or herhandwriting features. At first, this dissertation introduces the present research conditions of handwritingidentification and related theories, gives a survey of handwriting algorithm at home and abroad, and thenthis paper proposes the off-line HWI algorithm based on Haar lifting wavelet and SVM. The algorithmmainly includes three parts:
1. Pre-processing. Pre-processing in this paper mainly includes the following procedures: image grayprocessing, image binarization, noise removing, cutting words, normalization, texture formation and soon．
2. Feature extraction. Feature extraction in this paper is carried out based on texture map, and it isdivided into two parts: global texture feature extraction and single character texture feature extraction. Forglobal texture feature extraction, this paper proposes an algorithm based on two-dimensional Gabor filter.This algorithm uses32Kernel functions (4frequencies and8directions) for simulation training.32wavelettransform coefficients are gained through the convolution of texture, and then seek the variance value asglobal texture features. For single character texture feature extraction, this paper proposes the algorithmbased on Haar lifting wavelet. This algorithm, as second-generation wavelet transform, achieves DWT(discrete wavelet transforms) from integer to integer. Seek the variance of the wavelet coefficients throughdecomposing the single character sample into three levels and get the single character features. Aftercomprehensive analysis, the final results of the identification will be given.
3. Classifier design. This paper adopts the SVM classification algorithm to classify, including: SVMtraining and SVM classification. In the training phase, the input sample will be trained and the trainingresults will be saved; in the classification phase, inputting the test samples and specifying the trainingresults in order to achieve the classification of the test samples.
The experiment, selecting the handwriting samples of40individuals, is carried out in MATLAB7.0environment. Extract the texture features of the handwriting image by using2-dimensional Gabortransform and Haar lifting wavelet transform, and then complete the whole process of handwritingidentification after classifying by using the SVM classifier. The experiment achieves satisfactory results.
引文
[1]沈聪.基于改进的多通道Gabor小波变换的笔迹鉴别[D].北京：北京工业大学,2002.
[2] R．Plamondon,G Lorrete，Automatic Signature Verification and Writer Identification-the State of theArt,Pattern Recognition,vol.22,1989,PP.107~131
[3] V. Klement,An Application System for the Computer-assisted Identification of HandwritingInt.Camahan Conf.On Security Technology,1983,pp.75~79
[4] R．D．Naske, Writer Recognition by Prototype Related Deformation of Hand printed Characters，Proc.6th ICPR,l982,PP.819~822
[5] H．E．S．Said,T．N．Tan and K．D．Baker“,Personal Identification Based on Handwriting”．Proc．ofthe14th Inter．Conf．Pattern Recognition,Brisbane, PP.1761~1764,Australia,August1998．
[6]高雪铁.结合单字特征的笔迹鉴别研究[D].河北:河北工业大学,2008.
[7] H．E．S．Said,T．N．Tan and K．D．Baker, Personal Identification Based on Handwriting．Pattern Recognition,vol．33，no．1,2000,PP．149~160
[8] Yong Zhu,Tieniu Tan and Yunbong Wang, Biometric Personal Identification Based on Handwriting. ICPR2000. the15th International Conference on Pattern Recognition,2000,PP.801~804,Barcelona, Spain．
[9]田露.基于多特征数据融合的离线中文笔迹鉴别研究[D].河南:河南大学,2011.
[10] Cong Shen, Xiao-Gang Ruan, Tian-Lu Mao. Writer identification using gabor wavelet. InProceedings of the4th World Congress on Intelligent Control and Automation, Shanghai,P.R.China,2002：2061~2064.
[11] Kai Huang,Hong Yan. Off-line signature verification based on geometric feature extraction andneural network classification. Pattern Recognition,1997,30(1):9~17.
[12] Amit Jain,Aditya Kamat．Personal Identification Based on Handwriting，the paper，2000．
[13] Rafael C.Gonzalez,Richard E.Woods.数字图像处理[M].北京:电子工业出版社,2008.
[14]苗晓峰．基于纹理的文本依存的离线笔迹鉴别[D]．河北:河北工业大学,2006．
[15]范绪成．基于纹理特征提取的离线文字笔迹鉴别技术的研究[D]．浙江:浙江工业大学,2009．
[16] Kalle Karu, Anil K.. Jain, Ruudm, Bolle. Is there any texture in the image. Pattern Recognition,1996,29(9):1437~144.
[17] Rafael C.Gonzalez,Richard E.Woods,阮秋琦,阮宇智译.数字图像处理（第二版）[M].北京:电子工业出版社.2004.
[18]杨亚莉．基于纹理分析的笔迹鉴别方法研究[D]．武汉:武汉理工大学,2004.[62] Keerthi S S,CJ Lin. Asymptotic behaviors of support vector machines with Gaussian kernel[J].NeuralComputation,2003,15(7):1667-1689.
[19]张韵．基于纹理的文本独立离线笔迹鉴别[D]．河北:河北工业大学,2007．
[20]路娜．基于特征子空间的笔迹鉴别算法[D]．河北:河北工业大学,2009．
[21] P.Kruizinga, N.Petkov, S.E.Grigorescu. Comparison of texture features based on Gabor filters. InProceedings of the10th International Conference on Image Analysis and Processing, Venice, Italy,1999,142~147.
[22] Join M．Foley,Srinivasa Varadharajan,Chin C．Koh,Mylene C．Q．Farias．Detection of Gaborpatterns of different sizes， shapes，phases and eccentricities． Vision Research,47(2007)：85~107
[23] Linlin Shen,Li Bai． Mutual Boost learning for selecting Gabor features for facerecognition．Pattern Recognition Letters．27(2006)：1758~1767
[24] Mitja Perns，Horst Bischof,Chu Kiong Loo．Bio—computational model of object-recognition：Quantum Hebbian processing with neurally shaped Gabor wavelets．BioSystem82(2005)：116~126
[25] Ville Kyrki，Joni-Kristian Kamarainen，Heikki Kalviainen．Simple Gabor feature space forinvariant object recognition．Pattern Recognition Letters．25(2004)：311~318
[26] Laurenz Wiskott, Fellous Jean Marc, Norbert Kruger.et al. Face recongnition by elastic graphmatching. IEEE Transactions on Pattern Analysis and Machine Intelligence,1997,19(7):775~779.
[27]刘宏，李锦涛，崔国勤．基于SVM和纹理的笔迹鉴别方法．计算机辅助设计与图形学学报，2003，15(12)：1479～1484
[28]阮秋琦．数字图像处理学．北京：电子工业出版社，2001
[29] Stephane Mallat．A Wavelet Tour of Signal Processing(Second Edition)(杨力华，戴道消，黄文泉，湛秋辉译)．北京：机械工业出版社，2002：1~11
[30]李建平.小波分析与信号处理—理论、应用及软件实现.重庆:重庆出版社,1997.
[31] W.Sweldens. The Lifting Scheme:A Custom-design Construction of Biorthogonal Wavelets.Appl.Comput.Harmonic.Analysis,1996.3(2):186~200.
[32] W.Sweldens. The Lifting Scheme: Construction of Second-generation Wavelets. SIAMJ.Math.Anal.1998.29(2):411~446.
[33] S.Mallat. A theory for multi-resolution decomposition the wavelet representation. IEEE Trans PattnAnal Mach Intell,1989.11(7):674~693.
[34]孙延奎.小波分析及应用[M].北京:机械工业出版社,2005:100-116.
[35] Daubechies.W.Sweldens. Factoring Wavelet Transforms into Lifting Steps. Fourier Anal,1998,4(3):244~267.
[36]米晨,魏凤兰.Haar小波变换在图像处理中的应用.宁夏工程技术，2003,2（1）70~75.
[37]胡广书.现代信号处理教程.北京:清华人学出版社,2004.
[38] Theodoridis S, Koutroumbas K.Pattern Recognition[M].2nd.USA: El-sevier Science,2003.
[39]高学,金连文.一种基于支持向量机的手写汉字识别方法[J].电子学报,2002,30(5):13-20.
[40] Vladimirn.Vapnik,统计学习理论的本质[M].北京:清华大学出版社,2008:192-193.
[41]范劲松,方廷建.基于粗集理论和SVM算法的模式分类方法[J].模式识别与人工智能,2000,13（4）:419-423.
[42]崔伟东,周志华,李星.支持向量机研究[J].计算机工程与应用,2001,l:58-61.
[43]迪达.模式分类[M].北京:机械工业出版社,2003:252-261.
[44]桑金歌等.基于小训练样本和纹理分析的笔迹鉴别方法[J]．计算机教育,2008,2:122-124．
[45]王晓明.基于混合编程的笔迹鉴别系统的实现[D].天津:天津工业大学,2006.
[46]高雪铁.结合单字特征的笔迹鉴别算法[J].计算机与现代化,2010,3:133-137.
[47]马凤云等.SVM和BP相结合的垃圾邮件过滤技术[J].计算机安全,2006,6:32-34.
[48] Yang M H，Ahuja N．A Geometric Approach to Train Support Vector Machines[J]．IEEEProcessing of Computer Vision and PaRem Recognition，2000：430-437．
[49]古扎努尔·艾木肉拉.维吾尔文笔迹鉴别技术的研究与应用[D].北京:北京工业大学,2009.
[50]李毅等.应用支持向量机的纹理分类[J].通信学报,2005,
[51]张慧档.笔迹鉴别方法研究[D].郑州:郑州大学,2002.
[52]马立成.基于提升小波变换和HVS的半脆弱数字水印算法研究[D].上海:复旦大学,2007.
[53]陈倩倩.基于信息融合的笔迹鉴别算法研究[D].武汉:武汉理工大学,2010.
[54]谢承旺.不同种类支持向量机算法的比较研究[J].小型微型计算机系统.2008,1:106-109.
[55]张凯.支持向量机在汉字图像识别中的应用研究[D].合肥:合肥工业大学,2007.
[56]贾震斌.基于图像分析的汽车牌照定位和字符分割算法研究与实现[D].苏州:苏州大学,2006.
[57]方群会.暂态电能质量扰动检测及分类方法研究[D].重庆:重庆大学,2009.
[58]匡琳.支持向量机在文本分类中的应用的概述[J].科技资讯.2008,36:218-219.
[2] R．Plamondon,G Lorrete，Automatic Signature Verification and Writer Identification-the State of theArt,Pattern Recognition,vol.22,1989,PP.107~131
[3] V. Klement,An Application System for the Computer-assisted Identification of HandwritingInt.Camahan Conf.On Security Technology,1983,pp.75~79
[4] R．D．Naske, Writer Recognition by Prototype Related Deformation of Hand printed Characters，Proc.6th ICPR,l982,PP.819~822
[5] H．E．S．Said,T．N．Tan and K．D．Baker“,Personal Identification Based on Handwriting”．Proc．ofthe14th Inter．Conf．Pattern Recognition,Brisbane, PP.1761~1764,Australia,August1998．
[6]高雪铁.结合单字特征的笔迹鉴别研究[D].河北:河北工业大学,2008.
[7] H．E．S．Said,T．N．Tan and K．D．Baker, Personal Identification Based on Handwriting．Pattern Recognition,vol．33，no．1,2000,PP．149~160
[8] Yong Zhu,Tieniu Tan and Yunbong Wang, Biometric Personal Identification Based on Handwriting. ICPR2000. the15th International Conference on Pattern Recognition,2000,PP.801~804,Barcelona, Spain．
[9]田露.基于多特征数据融合的离线中文笔迹鉴别研究[D].河南:河南大学,2011.
[10] Cong Shen, Xiao-Gang Ruan, Tian-Lu Mao. Writer identification using gabor wavelet. InProceedings of the4th World Congress on Intelligent Control and Automation, Shanghai,P.R.China,2002：2061~2064.
[11] Kai Huang,Hong Yan. Off-line signature verification based on geometric feature extraction andneural network classification. Pattern Recognition,1997,30(1):9~17.
[12] Amit Jain,Aditya Kamat．Personal Identification Based on Handwriting，the paper，2000．
[13] Rafael C.Gonzalez,Richard E.Woods.数字图像处理[M].北京:电子工业出版社,2008.
[14]苗晓峰．基于纹理的文本依存的离线笔迹鉴别[D]．河北:河北工业大学,2006．
[15]范绪成．基于纹理特征提取的离线文字笔迹鉴别技术的研究[D]．浙江:浙江工业大学,2009．
[16] Kalle Karu, Anil K.. Jain, Ruudm, Bolle. Is there any texture in the image. Pattern Recognition,1996,29(9):1437~144.
[17] Rafael C.Gonzalez,Richard E.Woods,阮秋琦,阮宇智译.数字图像处理（第二版）[M].北京:电子工业出版社.2004.
[18]杨亚莉．基于纹理分析的笔迹鉴别方法研究[D]．武汉:武汉理工大学,2004.[62] Keerthi S S,CJ Lin. Asymptotic behaviors of support vector machines with Gaussian kernel[J].NeuralComputation,2003,15(7):1667-1689.
[19]张韵．基于纹理的文本独立离线笔迹鉴别[D]．河北:河北工业大学,2007．
[20]路娜．基于特征子空间的笔迹鉴别算法[D]．河北:河北工业大学,2009．
[21] P.Kruizinga, N.Petkov, S.E.Grigorescu. Comparison of texture features based on Gabor filters. InProceedings of the10th International Conference on Image Analysis and Processing, Venice, Italy,1999,142~147.
[22] Join M．Foley,Srinivasa Varadharajan,Chin C．Koh,Mylene C．Q．Farias．Detection of Gaborpatterns of different sizes， shapes，phases and eccentricities． Vision Research,47(2007)：85~107
[23] Linlin Shen,Li Bai． Mutual Boost learning for selecting Gabor features for facerecognition．Pattern Recognition Letters．27(2006)：1758~1767
[24] Mitja Perns，Horst Bischof,Chu Kiong Loo．Bio—computational model of object-recognition：Quantum Hebbian processing with neurally shaped Gabor wavelets．BioSystem82(2005)：116~126
[25] Ville Kyrki，Joni-Kristian Kamarainen，Heikki Kalviainen．Simple Gabor feature space forinvariant object recognition．Pattern Recognition Letters．25(2004)：311~318
[26] Laurenz Wiskott, Fellous Jean Marc, Norbert Kruger.et al. Face recongnition by elastic graphmatching. IEEE Transactions on Pattern Analysis and Machine Intelligence,1997,19(7):775~779.
[27]刘宏，李锦涛，崔国勤．基于SVM和纹理的笔迹鉴别方法．计算机辅助设计与图形学学报，2003，15(12)：1479～1484
[28]阮秋琦．数字图像处理学．北京：电子工业出版社，2001
[29] Stephane Mallat．A Wavelet Tour of Signal Processing(Second Edition)(杨力华，戴道消，黄文泉，湛秋辉译)．北京：机械工业出版社，2002：1~11
[30]李建平.小波分析与信号处理—理论、应用及软件实现.重庆:重庆出版社,1997.
[31] W.Sweldens. The Lifting Scheme:A Custom-design Construction of Biorthogonal Wavelets.Appl.Comput.Harmonic.Analysis,1996.3(2):186~200.
[32] W.Sweldens. The Lifting Scheme: Construction of Second-generation Wavelets. SIAMJ.Math.Anal.1998.29(2):411~446.
[33] S.Mallat. A theory for multi-resolution decomposition the wavelet representation. IEEE Trans PattnAnal Mach Intell,1989.11(7):674~693.
[34]孙延奎.小波分析及应用[M].北京:机械工业出版社,2005:100-116.
[35] Daubechies.W.Sweldens. Factoring Wavelet Transforms into Lifting Steps. Fourier Anal,1998,4(3):244~267.
[36]米晨,魏凤兰.Haar小波变换在图像处理中的应用.宁夏工程技术，2003,2（1）70~75.
[37]胡广书.现代信号处理教程.北京:清华人学出版社,2004.
[38] Theodoridis S, Koutroumbas K.Pattern Recognition[M].2nd.USA: El-sevier Science,2003.
[39]高学,金连文.一种基于支持向量机的手写汉字识别方法[J].电子学报,2002,30(5):13-20.
[40] Vladimirn.Vapnik,统计学习理论的本质[M].北京:清华大学出版社,2008:192-193.
[41]范劲松,方廷建.基于粗集理论和SVM算法的模式分类方法[J].模式识别与人工智能,2000,13（4）:419-423.
[42]崔伟东,周志华,李星.支持向量机研究[J].计算机工程与应用,2001,l:58-61.
[43]迪达.模式分类[M].北京:机械工业出版社,2003:252-261.
[44]桑金歌等.基于小训练样本和纹理分析的笔迹鉴别方法[J]．计算机教育,2008,2:122-124．
[45]王晓明.基于混合编程的笔迹鉴别系统的实现[D].天津:天津工业大学,2006.
[46]高雪铁.结合单字特征的笔迹鉴别算法[J].计算机与现代化,2010,3:133-137.
[47]马凤云等.SVM和BP相结合的垃圾邮件过滤技术[J].计算机安全,2006,6:32-34.
[48] Yang M H，Ahuja N．A Geometric Approach to Train Support Vector Machines[J]．IEEEProcessing of Computer Vision and PaRem Recognition，2000：430-437．
[49]古扎努尔·艾木肉拉.维吾尔文笔迹鉴别技术的研究与应用[D].北京:北京工业大学,2009.
[50]李毅等.应用支持向量机的纹理分类[J].通信学报,2005,
[51]张慧档.笔迹鉴别方法研究[D].郑州:郑州大学,2002.
[52]马立成.基于提升小波变换和HVS的半脆弱数字水印算法研究[D].上海:复旦大学,2007.
[53]陈倩倩.基于信息融合的笔迹鉴别算法研究[D].武汉:武汉理工大学,2010.
[54]谢承旺.不同种类支持向量机算法的比较研究[J].小型微型计算机系统.2008,1:106-109.
[55]张凯.支持向量机在汉字图像识别中的应用研究[D].合肥:合肥工业大学,2007.
[56]贾震斌.基于图像分析的汽车牌照定位和字符分割算法研究与实现[D].苏州:苏州大学,2006.
[57]方群会.暂态电能质量扰动检测及分类方法研究[D].重庆:重庆大学,2009.
[58]匡琳.支持向量机在文本分类中的应用的概述[J].科技资讯.2008,36:218-219.