图像中字符识别算法的设计与实现
|
摘要
借助于数学理论的研究和进步以及计算机技术的发展,数字图像处理技术越来越多的应用到各个领域。模式识别通过用机器代替人眼对未知事物进行判断,具有较高应用价值,因而成为图像处理领域中的重要分支。字符识别技术由于具有广阔的应用前景,得到了快速的发展,至今为止,已经成功运用于OCR以及车牌识别中。然而,与具体工作场景相关联、要满足具体要求的字符识别,具有一
定难度,仍然处于研究探索阶段。
本文中的标牌字符识别子系统包括原始图像的预处理、椭圆标牌定位、字符区域的提取、字符的分割、字符识别几个过程。
图像预处理中,通过分析背景信息,对灰度化的图像使用全局阈值法分割得到二值图像,并根据实际情况将背景分为几种类型。用形态学方法去除小的连通区域,结合椭圆特征去除另外的干扰区域。
在椭圆标牌的定位和字符分割部分,使用最小二乘拟合方法对椭圆边界进行拟合,得到椭圆几何参数,包括椭圆中心点坐标,长短轴长度以及倾斜角度。根据霍夫变换检测到的直线斜率对图像进行旋转,根据椭圆拟合得到的几何参数对图像进行错切以及缩放变换。经过这一系列的几何变换,得到了近似正圆区域。利用椭圆中心位置以及椭圆形状特征分割出矩形字符区域。分析投影法字符分割法的优缺点,用投影法结合先验知识分割字符。
在字符识别部分中,讨论了几种特征值的选取方法,分析了各自优缺点。对于带有惩罚因子的模板匹配方法,提出了选择连通背景区域中过型心的水平直线上到两边字符区域的线段中点作为惩罚点的方法。设计实现了以类间散步矩阵为产生矩阵的主成分分析字符识别算法。对于BP神经网络识别方法,设计了输入和输出数据格式,确定了输入输出层神经元个数、传递函数,试验选取了合适的隐层神经元个数。用样本数据测试各个模式识别算法,分析对比识别结果,提出了同时运用两种识别方法进行识别从而提高结果可信度的识别方法。
Benefiting from the research and progress of mathematical theories and the development of computer science, digital image processing technology is more and more widely used in various areas. With the ability to learn unknown entities through vehicles instead of human eyes, pattern recognition has great application perspects and becomes one important branch in image processing field. Due to wide application prospects, character recognition has been developing quickly, and is practically used in OCR and vehicle license plate recognition fields now. Whereas, the character recognition, which abides a certain background and must meet a desired goal, is difficult and is still under research.
In this thesis, the plate character recognition sub system consists of original image preprocessing, elliptical plate positioning, extraction of character region, character segmentation and character recognition.
In the process of image processing, global threshold method is applied to the gray image to get binary image, classify the binary image to several conditions due to concrete situation, morphological operation is used to eliminate small connected regions and other useless regions.
In the process of elliptical plate extraction and character segmentation, direct least squared fitting method is applied to the edge of the ellipse to get geometric parameters of the ellipse, including coordinates of the centre point, the lengths of the long-axis and short-axis. Rotate the image using the line slope detecting by Hough transformation, do shear transformation and resizing transformation with geometric parameters getting by ellipse fitting. Get an approximate circular area as a result of a collection of geometric transformations. Extract rectangular character region by center of the ellipse and the shape characteristics. Analyze the merit and demerit of projection segmentation and segment each character by projection method while considering some prior knowledge.
In the process of character recognition, several methods of eigen value selection is discussed and merit and demerit of each one is analyzed. In terms of template matching with penalty points method, selecting the middle point of the line segment of the line passing through the center point of the connected regions in the background which ends by the character region on two ends, as the penalty points is proposed. The principal component analysis recognition algorithm with intraclass scatter matrix as the generating matrix is designed and implemented. For BP neutral network, data format of input and output is designed, the number of input and output neurons, transfer function are determined, proper number of hidden layer neurons is selected though experiment. Each pattern recognition algorithm is tested by the sample data, analyze failed cases. One method which combines two pattern recognition methods with high success rate to make the result more reliable is proposed after analyzing and comparing the experiment result.
定难度,仍然处于研究探索阶段。
本文中的标牌字符识别子系统包括原始图像的预处理、椭圆标牌定位、字符区域的提取、字符的分割、字符识别几个过程。
图像预处理中,通过分析背景信息,对灰度化的图像使用全局阈值法分割得到二值图像,并根据实际情况将背景分为几种类型。用形态学方法去除小的连通区域,结合椭圆特征去除另外的干扰区域。
在椭圆标牌的定位和字符分割部分,使用最小二乘拟合方法对椭圆边界进行拟合,得到椭圆几何参数,包括椭圆中心点坐标,长短轴长度以及倾斜角度。根据霍夫变换检测到的直线斜率对图像进行旋转,根据椭圆拟合得到的几何参数对图像进行错切以及缩放变换。经过这一系列的几何变换,得到了近似正圆区域。利用椭圆中心位置以及椭圆形状特征分割出矩形字符区域。分析投影法字符分割法的优缺点,用投影法结合先验知识分割字符。
在字符识别部分中,讨论了几种特征值的选取方法,分析了各自优缺点。对于带有惩罚因子的模板匹配方法,提出了选择连通背景区域中过型心的水平直线上到两边字符区域的线段中点作为惩罚点的方法。设计实现了以类间散步矩阵为产生矩阵的主成分分析字符识别算法。对于BP神经网络识别方法,设计了输入和输出数据格式,确定了输入输出层神经元个数、传递函数,试验选取了合适的隐层神经元个数。用样本数据测试各个模式识别算法,分析对比识别结果,提出了同时运用两种识别方法进行识别从而提高结果可信度的识别方法。
Benefiting from the research and progress of mathematical theories and the development of computer science, digital image processing technology is more and more widely used in various areas. With the ability to learn unknown entities through vehicles instead of human eyes, pattern recognition has great application perspects and becomes one important branch in image processing field. Due to wide application prospects, character recognition has been developing quickly, and is practically used in OCR and vehicle license plate recognition fields now. Whereas, the character recognition, which abides a certain background and must meet a desired goal, is difficult and is still under research.
In this thesis, the plate character recognition sub system consists of original image preprocessing, elliptical plate positioning, extraction of character region, character segmentation and character recognition.
In the process of image processing, global threshold method is applied to the gray image to get binary image, classify the binary image to several conditions due to concrete situation, morphological operation is used to eliminate small connected regions and other useless regions.
In the process of elliptical plate extraction and character segmentation, direct least squared fitting method is applied to the edge of the ellipse to get geometric parameters of the ellipse, including coordinates of the centre point, the lengths of the long-axis and short-axis. Rotate the image using the line slope detecting by Hough transformation, do shear transformation and resizing transformation with geometric parameters getting by ellipse fitting. Get an approximate circular area as a result of a collection of geometric transformations. Extract rectangular character region by center of the ellipse and the shape characteristics. Analyze the merit and demerit of projection segmentation and segment each character by projection method while considering some prior knowledge.
In the process of character recognition, several methods of eigen value selection is discussed and merit and demerit of each one is analyzed. In terms of template matching with penalty points method, selecting the middle point of the line segment of the line passing through the center point of the connected regions in the background which ends by the character region on two ends, as the penalty points is proposed. The principal component analysis recognition algorithm with intraclass scatter matrix as the generating matrix is designed and implemented. For BP neutral network, data format of input and output is designed, the number of input and output neurons, transfer function are determined, proper number of hidden layer neurons is selected though experiment. Each pattern recognition algorithm is tested by the sample data, analyze failed cases. One method which combines two pattern recognition methods with high success rate to make the result more reliable is proposed after analyzing and comparing the experiment result.
引文
[1] Pratt K. Digital Image Processing: [Masters Thesis]. New York: John Wiley, 1978
[2] O‘Handley D A, Green W B. Recent Development in Digital Image Processing at the Image Processing Laboratory at the Jet Propulsion Laboratory. Proc. IEEE, 1972, 60 (7):821-828
[3]房建成,万德钧.GPS组合导航系统在车辆导航的应用.东南大学学报,1996,26(3):97-101
[4]方珉,刘星荣.GPS技术惯性导航技术和组合导航技术在汽车导航中的应用.汽车技术,2000,5:1-3
[5]王维,张英,任国全.一种新型高精度地磁定向方法研究与系统设.传感技术学报,2009,22(1):143-145
[6]山西省煤炭地质公司.陀螺经纬仪3种定向方法的研究与探讨.科技情报开发与经济,2009,19(34):155-157
[7] Schyndel R G, Tirkel A Z, Osborne C F. A Digital Watermark. Proceedings of IEEE International Conference on Image Processing, 1994, 2:86-90
[8]肖毅,步金梅,等.图像的数字化处理及其文件格式特点.中国新技术新产品,2009,5:12-13
[9]姚敏.数字图像处理.北京:机械工业出版社,2006,119-122
[10]霍洪涛.数字图像处理.北京:机械工业出版社,2003,3-5
[11]蒋先刚.数字图像模式识别工程软件设计.北京:中国水利水电出版社,2008,21-22
[12] Salwo W, Jerigan M E, Robert D D. Comparison of Color Image Edge Detectors In Multiple Color Spaces. IEEE International Conference Conference on Image processing, 2000, 2:796-799
[13] Ethan E D, Karen A P, Sos S A. Feature Extraction System for Contextual Classification within Security Imaging Applications. IEEE International Conference on System of Systems Engineering, 2007, 6:1-6
[14] Andreas K. A Comparative Study on Color Edge Detection. Asian Conference on Computer Vision ACCV’95, 1995, 3:547-478
[15]白俊江,洪春勇.基于Sobel的边缘检测方法.电脑知识与技术,2009,5(21):5847-5849
[16]雷丽珍.数字图像边缘检测方法的探讨.测绘通报,2006,3:40-42
[17] Slawomir B W. Color Image Edge Detection and Segmentation: A comparison of theVector Angle and Euclidean Distance Color Similarity Measures:[Masters Thesis]. Waterloo, Canada:University of Waterloo, 1999
[18] Naito T, Tsukada T, Yamada K, Kozuka K, Yamamoto S. License Plate Recognition Method for Inclined Plates Outdoors. Proceedings of the 1999 International Conference on Information Intelligence and Systems, 1999:304-312
[19] Naito T, Tsukada T, Yamada K, Kozuka K, Yamamoto S. Robust License-Plate Recognition Method for Passing Vehicles under Outside Environment.IEEE Transaction on Vehicular Technology, 2000, 49(6):2309-2319
[20] Yamada K, Nakano T, Yamamoto S. A Vision Sensor Having an Expanded Dynamic Range for Autonomous Vehicles. IEEE Transaction on Vehicular Technology, 1998, 47(1):332-341
[21]郑文明.基于K-L变换的车辆牌照自动识别技术研究与实现:[硕士学位论文].泉州:华侨大学,2001
[22]陈亮,杜宇人.基于字符边缘检测和颜色特征的车牌定位方法.扬州大学学报,2008,11(3):57-60
[23]任彬,汪炳权.基于直方图指数平滑的闭值和峰点自动检测方法.中国图象图形学报,1997,4:230-233.
[24]张辉,张道勇.灰度等级处理中的OSTU动态阈值法研究.研究与开发,2008,7:24-27
[25]李春娟,赵艳花.一种基于形态学和几何特征的车牌定位方法.模式识别,2009,25(11-1):175-177
[26]张云刚,张长水.利用Hough变换和先验知识的车牌字符分割算法.计算机学报,2004,(1):130-135
[27]杨全银,魏雅杰,等.基于Hough变换的线段检测算法.红外与激光工程,2009,38(3):525-527
[28]费旭东,王文剑.二值图边界的快速提取.信号处理,1994,10(3):130-137
[29] Andrew F, Maurizio P. Direct Least Square Fitting of Ellipses. IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE, 1999, 21(5):477-480
[30]张德丰.数字图像处理(Matlab版).北京:人民邮电出版社,2009,130-132
[31]杨磊,邓天民.车辆牌照自动识别系统中的字符分割技术.重庆交通大学学报,2008,27(1):997-998
[32]焦婷婷,侯德文.车牌字符分割问题分析.山东科技,2008,21(2):39-42
[33]刘培锋,张文斌.灰度图像下车牌定位与分割算法研究.微计算机信息,2008,24(6-3):281-283
[34]杨枝灵.Visua1C++数字图像获取、处理及实践应用.北京:人民邮电出版社,2003,570-581
[35]宋日聪,胡伟.手写体数字识别系统中一种新的特征提取方案.计算机科学,2007,34(9):236-238
[36]凌彤辉.车辆牌照自动识别系统的研究:[硕士学位论文].成都:四川大学,2005
[37]陈祥光.神经网络技术及应用.北京:中国电力出版社,2003,19-126
[2] O‘Handley D A, Green W B. Recent Development in Digital Image Processing at the Image Processing Laboratory at the Jet Propulsion Laboratory. Proc. IEEE, 1972, 60 (7):821-828
[3]房建成,万德钧.GPS组合导航系统在车辆导航的应用.东南大学学报,1996,26(3):97-101
[4]方珉,刘星荣.GPS技术惯性导航技术和组合导航技术在汽车导航中的应用.汽车技术,2000,5:1-3
[5]王维,张英,任国全.一种新型高精度地磁定向方法研究与系统设.传感技术学报,2009,22(1):143-145
[6]山西省煤炭地质公司.陀螺经纬仪3种定向方法的研究与探讨.科技情报开发与经济,2009,19(34):155-157
[7] Schyndel R G, Tirkel A Z, Osborne C F. A Digital Watermark. Proceedings of IEEE International Conference on Image Processing, 1994, 2:86-90
[8]肖毅,步金梅,等.图像的数字化处理及其文件格式特点.中国新技术新产品,2009,5:12-13
[9]姚敏.数字图像处理.北京:机械工业出版社,2006,119-122
[10]霍洪涛.数字图像处理.北京:机械工业出版社,2003,3-5
[11]蒋先刚.数字图像模式识别工程软件设计.北京:中国水利水电出版社,2008,21-22
[12] Salwo W, Jerigan M E, Robert D D. Comparison of Color Image Edge Detectors In Multiple Color Spaces. IEEE International Conference Conference on Image processing, 2000, 2:796-799
[13] Ethan E D, Karen A P, Sos S A. Feature Extraction System for Contextual Classification within Security Imaging Applications. IEEE International Conference on System of Systems Engineering, 2007, 6:1-6
[14] Andreas K. A Comparative Study on Color Edge Detection. Asian Conference on Computer Vision ACCV’95, 1995, 3:547-478
[15]白俊江,洪春勇.基于Sobel的边缘检测方法.电脑知识与技术,2009,5(21):5847-5849
[16]雷丽珍.数字图像边缘检测方法的探讨.测绘通报,2006,3:40-42
[17] Slawomir B W. Color Image Edge Detection and Segmentation: A comparison of theVector Angle and Euclidean Distance Color Similarity Measures:[Masters Thesis]. Waterloo, Canada:University of Waterloo, 1999
[18] Naito T, Tsukada T, Yamada K, Kozuka K, Yamamoto S. License Plate Recognition Method for Inclined Plates Outdoors. Proceedings of the 1999 International Conference on Information Intelligence and Systems, 1999:304-312
[19] Naito T, Tsukada T, Yamada K, Kozuka K, Yamamoto S. Robust License-Plate Recognition Method for Passing Vehicles under Outside Environment.IEEE Transaction on Vehicular Technology, 2000, 49(6):2309-2319
[20] Yamada K, Nakano T, Yamamoto S. A Vision Sensor Having an Expanded Dynamic Range for Autonomous Vehicles. IEEE Transaction on Vehicular Technology, 1998, 47(1):332-341
[21]郑文明.基于K-L变换的车辆牌照自动识别技术研究与实现:[硕士学位论文].泉州:华侨大学,2001
[22]陈亮,杜宇人.基于字符边缘检测和颜色特征的车牌定位方法.扬州大学学报,2008,11(3):57-60
[23]任彬,汪炳权.基于直方图指数平滑的闭值和峰点自动检测方法.中国图象图形学报,1997,4:230-233.
[24]张辉,张道勇.灰度等级处理中的OSTU动态阈值法研究.研究与开发,2008,7:24-27
[25]李春娟,赵艳花.一种基于形态学和几何特征的车牌定位方法.模式识别,2009,25(11-1):175-177
[26]张云刚,张长水.利用Hough变换和先验知识的车牌字符分割算法.计算机学报,2004,(1):130-135
[27]杨全银,魏雅杰,等.基于Hough变换的线段检测算法.红外与激光工程,2009,38(3):525-527
[28]费旭东,王文剑.二值图边界的快速提取.信号处理,1994,10(3):130-137
[29] Andrew F, Maurizio P. Direct Least Square Fitting of Ellipses. IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE, 1999, 21(5):477-480
[30]张德丰.数字图像处理(Matlab版).北京:人民邮电出版社,2009,130-132
[31]杨磊,邓天民.车辆牌照自动识别系统中的字符分割技术.重庆交通大学学报,2008,27(1):997-998
[32]焦婷婷,侯德文.车牌字符分割问题分析.山东科技,2008,21(2):39-42
[33]刘培锋,张文斌.灰度图像下车牌定位与分割算法研究.微计算机信息,2008,24(6-3):281-283
[34]杨枝灵.Visua1C++数字图像获取、处理及实践应用.北京:人民邮电出版社,2003,570-581
[35]宋日聪,胡伟.手写体数字识别系统中一种新的特征提取方案.计算机科学,2007,34(9):236-238
[36]凌彤辉.车辆牌照自动识别系统的研究:[硕士学位论文].成都:四川大学,2005
[37]陈祥光.神经网络技术及应用.北京:中国电力出版社,2003,19-126
© 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 |