基于图像分析的第二代身份证字符识别技术研究
|
摘要
图像字符识别是图像处理与模式识别理论的一个重要应用领域,是实现智能人机接口的重要途径;近几十年来得到广泛的研究。字符识别是模式识别的一个重要方面,在信息处理,办公自动化,邮政系统,银行系统等方面有着重要的使用价值和理论意义。
目前文字识别技术已相对成熟,但是对粘连字符识别错误率还比较高。一般认为主要原因是粘连字符的错误切分而导致字符的严重失真变形,无法正确的识别,因此粘连字符的切分成为提高识别率的关键技术。现有的切分方法主要有:(1)基于图像分析的直接切分法,通过图像分析寻找字符之间较为合理的切分点,但切分错误率比较高;(2)基于识别的切分法,先通过图像分析,确定几个可能的切分点,借助识别结果,选择合理的切分点。后者切分方法的识别比较高,但是多次识别,步骤繁琐,比较耗时。
为了提高粘连字符识别率与识别速度,本文的研究与创新点可分为两大部分:
一是在第二代身份证识别算法方面,包括三点:(1)适合第二代身份证的感兴趣区域提取的分割方法,实现文字区域的定位和分割,将真正的文字图形从身份证复杂的背景中分割出来。运用一定的识别原理,对文字进行分类,确定其属性;(2)第二代身份证的字符切分方法,降低切分字符的粘连率,针对污点、褪色、光照不均匀和分辨率过低对检测区域的字符粘连影响采用了上下轮廓凹凸特征近似检测单个字符的宽度,在字符宽度的约束下,根据轮廓凹凸特征,直接建立切分路径,提高文字字符的识别正确率;(3)针对第二代身份证图像字符识别,提出了一种基于字符串轮廓检测的方法。该方法采用对图像字符区域特征提取后,获取字符区域后对其上下轮廓凹凸特征近似检测单个字符的宽度,从中选出稳定的局部特征,利用结构语句识别的方法进行字符识别。
二是设计实现第二代身份证识别系统软件原型,字符识别系统主要分三个部分:(1)预处理模块:这个模块不仅包含了本文所采用的算法先对图像进行灰度拉伸,在灰度拉伸的基础上对图像采用空间域法进行图像的滤波。而且还包含了本文第三章所采用的对图像进行字符区域分割。(2)特征提取模块:采用上下轮廓凹凸特征近似检测单个字符的宽度,在字符宽度的约束下,根据轮廓凹凸特征,对字符进行特征的提取。(3)字符识别模块:采用字符轮廓结构特征和统计特征相结合的方法,并将字符库中的字符分为纹理图和非纹理图,然后分别对纹理图和非纹理图采用区域匹配,达到匹配的目的。
Image character identification is an important field of application in image processing and Pattern Recognition theory, and it is the major way of realizing intelligent human-computer communication. In recent dozens of years, they are extensively researched. Character identification is an important part of Pattern Recognition, and it has main useful value and theory significance in information processing, office automation, post system, bank system.
At the present time, OCR(Optical Character Recognition)character identification's technique has already relatively matured, but the error rate of conglutination character identification is rather high. Generally speaking, the main reason is that the error syncopate of the conglutination character can lead to character's severity distortion deformation, so correct identification can't be obtained, therefore syncopate of conglutination character identification become a key technique to improve identification rate. Existing syncopate methods are: (1)the direct syncopate method based image analysis: the reasonable syncopate point among the characters is looked for through image analysis, but the error of syncopate is rather high; (2)the syncopate method based identification: several possible syncopate points are determined through image analysis, and reasonable syncopate point is chose by means of the result of identification. The latter syncopate method's identification is higher than the first one, but it needs to be identified many times, the step is verbose and time-consuming.
To improve the identification rate of conglutination character and the speed of identification, this thesis's main research and innovations are divided into two parts:
First, aiming at the algorithm of the second generation ID card's character identification, three points are included: (l)The method of segmentation which is adequate for extracting interested areas of the second generation ID card(The third chapter), to realize location and segmentation of characters, the real character image is segmented from complicated background of ID card, and some unessential signal are disposed according to certain norm, and the size of characters is normalized , position and font-weight of stroke, and certain identification principle is used to classify the character, its property is determined; (2) The character syncopate method of the second generation ID card: to reduce conglutination rate of syncopate character, and aiming at stain, fade, asymmetrical illumination and rather low resolution which can effect detection region's conglutination character, up and down profile concavo-convex feature are adopted to approximately detect single character's width, the restrict condition of character's width is used to directly set up syncopate path according to profile concavo-convex feature, so the correct rate of character identification can be improved; (3) Aiming at the second generation ID card's character identification, a method which is put forward to detect character string's profile. Feature is firstly extracted from image character region in this method(The fifth chapter), after the areas of character are gained, and then fluctuation profile concavo-convex character is adopted to approximately detect single character's width, and the steady local character is accordingly chose, and frame sentence identification is used to identify the character.
Second, software prototype is designed to achieve the second generation ID card's identification system(The sixth chapter), three parts are divided by character identification system: (1)Pretreatment module: algorithm is adopted in this module in the thesis, the image is stretched by gray, and then the image is filtered by the method of space domain, which is based on gray stretching, and the method of character segmentation is adopted in the third chapter in this thesis. (2)Features extracted module: up and down profile concavo-convex feature is adopted to approximately detect single character's width, the character is extracted under the restrict condition of character's width and according to profile concavo-convex feature. (3)Character identification module: a method which combines structure feature of character profile with statistics feature is adopted, and the characters from character library are divided into texture image and non-texture image, and then the texture image and non-texture image are matched by region, so the purpose of matching is achieved.
目前文字识别技术已相对成熟,但是对粘连字符识别错误率还比较高。一般认为主要原因是粘连字符的错误切分而导致字符的严重失真变形,无法正确的识别,因此粘连字符的切分成为提高识别率的关键技术。现有的切分方法主要有:(1)基于图像分析的直接切分法,通过图像分析寻找字符之间较为合理的切分点,但切分错误率比较高;(2)基于识别的切分法,先通过图像分析,确定几个可能的切分点,借助识别结果,选择合理的切分点。后者切分方法的识别比较高,但是多次识别,步骤繁琐,比较耗时。
为了提高粘连字符识别率与识别速度,本文的研究与创新点可分为两大部分:
一是在第二代身份证识别算法方面,包括三点:(1)适合第二代身份证的感兴趣区域提取的分割方法,实现文字区域的定位和分割,将真正的文字图形从身份证复杂的背景中分割出来。运用一定的识别原理,对文字进行分类,确定其属性;(2)第二代身份证的字符切分方法,降低切分字符的粘连率,针对污点、褪色、光照不均匀和分辨率过低对检测区域的字符粘连影响采用了上下轮廓凹凸特征近似检测单个字符的宽度,在字符宽度的约束下,根据轮廓凹凸特征,直接建立切分路径,提高文字字符的识别正确率;(3)针对第二代身份证图像字符识别,提出了一种基于字符串轮廓检测的方法。该方法采用对图像字符区域特征提取后,获取字符区域后对其上下轮廓凹凸特征近似检测单个字符的宽度,从中选出稳定的局部特征,利用结构语句识别的方法进行字符识别。
二是设计实现第二代身份证识别系统软件原型,字符识别系统主要分三个部分:(1)预处理模块:这个模块不仅包含了本文所采用的算法先对图像进行灰度拉伸,在灰度拉伸的基础上对图像采用空间域法进行图像的滤波。而且还包含了本文第三章所采用的对图像进行字符区域分割。(2)特征提取模块:采用上下轮廓凹凸特征近似检测单个字符的宽度,在字符宽度的约束下,根据轮廓凹凸特征,对字符进行特征的提取。(3)字符识别模块:采用字符轮廓结构特征和统计特征相结合的方法,并将字符库中的字符分为纹理图和非纹理图,然后分别对纹理图和非纹理图采用区域匹配,达到匹配的目的。
Image character identification is an important field of application in image processing and Pattern Recognition theory, and it is the major way of realizing intelligent human-computer communication. In recent dozens of years, they are extensively researched. Character identification is an important part of Pattern Recognition, and it has main useful value and theory significance in information processing, office automation, post system, bank system.
At the present time, OCR(Optical Character Recognition)character identification's technique has already relatively matured, but the error rate of conglutination character identification is rather high. Generally speaking, the main reason is that the error syncopate of the conglutination character can lead to character's severity distortion deformation, so correct identification can't be obtained, therefore syncopate of conglutination character identification become a key technique to improve identification rate. Existing syncopate methods are: (1)the direct syncopate method based image analysis: the reasonable syncopate point among the characters is looked for through image analysis, but the error of syncopate is rather high; (2)the syncopate method based identification: several possible syncopate points are determined through image analysis, and reasonable syncopate point is chose by means of the result of identification. The latter syncopate method's identification is higher than the first one, but it needs to be identified many times, the step is verbose and time-consuming.
To improve the identification rate of conglutination character and the speed of identification, this thesis's main research and innovations are divided into two parts:
First, aiming at the algorithm of the second generation ID card's character identification, three points are included: (l)The method of segmentation which is adequate for extracting interested areas of the second generation ID card(The third chapter), to realize location and segmentation of characters, the real character image is segmented from complicated background of ID card, and some unessential signal are disposed according to certain norm, and the size of characters is normalized , position and font-weight of stroke, and certain identification principle is used to classify the character, its property is determined; (2) The character syncopate method of the second generation ID card: to reduce conglutination rate of syncopate character, and aiming at stain, fade, asymmetrical illumination and rather low resolution which can effect detection region's conglutination character, up and down profile concavo-convex feature are adopted to approximately detect single character's width, the restrict condition of character's width is used to directly set up syncopate path according to profile concavo-convex feature, so the correct rate of character identification can be improved; (3) Aiming at the second generation ID card's character identification, a method which is put forward to detect character string's profile. Feature is firstly extracted from image character region in this method(The fifth chapter), after the areas of character are gained, and then fluctuation profile concavo-convex character is adopted to approximately detect single character's width, and the steady local character is accordingly chose, and frame sentence identification is used to identify the character.
Second, software prototype is designed to achieve the second generation ID card's identification system(The sixth chapter), three parts are divided by character identification system: (1)Pretreatment module: algorithm is adopted in this module in the thesis, the image is stretched by gray, and then the image is filtered by the method of space domain, which is based on gray stretching, and the method of character segmentation is adopted in the third chapter in this thesis. (2)Features extracted module: up and down profile concavo-convex feature is adopted to approximately detect single character's width, the character is extracted under the restrict condition of character's width and according to profile concavo-convex feature. (3)Character identification module: a method which combines structure feature of character profile with statistics feature is adopted, and the characters from character library are divided into texture image and non-texture image, and then the texture image and non-texture image are matched by region, so the purpose of matching is achieved.
引文
[1]阮秋琦.数字图像处理学,电子工业出版社,2007,412-420.
[2]Fumera G,Roli F,A Theoretical and Experimental Analysis of Linear Combinersfor Multiple Classifier Systems.Pattern Analysis and Machine Intelligence,IEEE Transactions,2005,27(6):942-956
[3]Lienhart R,Wernicke A.Localizing and segmenting text in images and videos.IEEE Trans.Circuits and Systems for Video Technology,2002,12(4):256-268.
[4]胡小锋,周勇,叶庆泰.复杂背景彩色图像中的文字分割.上海交通大学学报,2006,141-147.
[5]叶晨洲,杨杰,宣国荣.车辆牌照字符识别.上海交通大学学报,2000,672-675.
[6]王学文,丁晓青,刘长松.基于Gabor变换的高鲁棒汉字识别新方法.电子学报,2002.9:1317-1321.
[7]胡小锋,赵辉.Visual c++/MATLAB图像处理与识别实用案例精选.人民邮电出版社,2004,194-253.
[8]Li C,Ding X Q,Wu Y S.Automitic text segmentation and text recognition for video indexing.Multimedia Systems,2000,8(1):69-81.
[9]李一舒.车牌字符识别的改进算法研究:(硕士学位论文).杭州:浙江大学,2007.
[10]张引,潘云鹤.彩色汽车图像牌照定位新方法.中国图像图形学报,2001,6(4):374-377.
[11]Louka Dlagnekov.Video-based Car Surveillance:Licenes Plate,Make,and Model Recognition.CSD Master's Thesis 2005.
[12]张宏烈.支持向量机在字符识别中的应用研究.微计算机信息,2006,11:245-247.
[13]孙忠贵,王玲.数字图像直方图均衡化的自适应校正研究.计算机时代.2004,11:19-20.
[14]Cormen,T.H.Introduction to Algorithm.机械工业出版社,2006.
[15]Chen Datong,Odobez Jean-Marc,Bourlard Herve.Text detection and recognition in images and video frames.Pattern Recognition.2004,37(3):595-608.
[16]王建平,钱自拓.基于数学形态学的图像汉字笔划细化和提取.合肥工业大学学报,2005,11(28):1431-1435.
[17]李明.复杂背景下的文本定位与提取研究:(硕士学位论文).合肥:合肥工业大学,2004.
[18]许剑锋.数字视频中的文本分割的研究:(博士学位论文).广州:华南理工大学,2005.
[19]王勇,郑辉,胡德文.图像和视频中的文字获取技术.中国图像图形学报,2004,5(9):532-537.
[20]Rafael C G,Richard E W.Digital Image Processing Second Edition.BeiJing,Publish House of Electronics Industry,2004,23:150-204.
[21]Lee C N,Wu B.Chinese-character-stroke-extraction algorithm based on contour information.Pattern Recognition,1998,31(6):651-663.
[22]刘峡壁,贾云得.汉字笔段形成规律及其提取方法.计算机学报,2004,27(3):390-395.
[23]孙星明,杨茂江,刘国华.完全基于结构知识的汉字笔画抽取方法.计算机研究与发展,2000,37(5):543-550.
[24]章毓晋.图像分割.北京:科学出版社,2001.
[25]崔屹.图像处理与分析——数学形态学方法及应用.北京:科学出版社,2002,5:23-34.
[26]Yi Lu,B.Harmon,J.Trenkle and R.Vogt,"An accurate and efficient system for segmenting machine-printed text" U.S Postal Service 5th Advanced Technology Conference,WashingtonD.C.November,1992,3:93-105.
[27]石洪贵.基于机器视觉的字符识别技术研究:(硕士学位论文).合肥:合肥工业大学,2006.
[28]刘倩.基于彩色图像的文本区域提取研究:(硕士学位论文).成都:西南交通大学,2003.
[29]章东平.视频文本的提取:(博士学位论文).杭州:浙江大学,2006.
[30]王学文,丁晓青,刘长松.基于Gabor变换的高鲁棒汉字识别新方法,电子学报,2002,9:1317-1322.
[31]陈芳信,马翔,姜新等.基于多神经网络的手写数字综合识别系统.信息技术,2003,27(9):1-4.
[32]孙亦南,刘伟军,王越超.基于儿何不变量的图像特征识别.计算机应用与软件,2004,12,21(12):1-3.
[33]孙即祥.模式识别中的特征提取与计算机视觉不变量.北京:国防工业出版社,2001.
[34]Zhe Chen.Extracting Projective Invariant of 3D Line Set from a SingleView.IEEE International Conference on Image Processing,2004,33:2119-2122.
[35]Sugimoto A.Geometric invariant of noncoplanar lines in a single view.Proceedings of International Conference on Pattern Recognition,Jerusalem,Israel 1994,5:190-195.
[36]Senem Velipasalar,Wayne Wolf.Recovering Field of View Lines by Using Projective Invariants.IEEE International Conference on Image Processing,2004,32:3069-3072.
[37]夏良正.数字图像处理.东南大学出版社,1999.
[38]容观澳.计算机图像处理.清华大学出版社,2000.
[39]R.M.Haralick,K.Shanmugam,I.Dinstein.Texture Feature for Image Classification,IEEE Trans.Systems Man Cybernet.1973,3(1):610-621.
[40]M.M.Galloway.Texture Classification Using Gray-level Run lengths,CLIP 4,1975,44:172-179.
[41]苏厚胜.车牌识别系统的设计与实现.可编程控制与工厂自动化,2005,3(3):78-90.
[42]孟国强.车辆车牌识别系统的原理及算法研究.计算机应用,2001,15(4):548-570.
[43]冯桂.图像直方图不变特性在影像匹配定位中的应用.计算机辅助设计与图形学学报,2000,12(2):144-200.
[44]蒋伟峰,刘济林.基于改进型CLAFIC学习子空间算法的有限汉字集识别.软件学报,2000,7:679-684.
[45]阎建国,高亮,卢京潮.图象处理技术在车牌识别中的应用.计算机应用,2000,22(1):17-18.
[46]楼兵军.基于区域的图像匹配算法的关键技术研究:(硕士学位论文).西安:西安电子科技大学,2006.
[47]孟丽娜,田玉敏.基于区域的图像检索技术研究:(硕士学位论文).西安:西安电子科技大学,2005.
[48]G.R.Cross,A.K.Jain,Markov random field texture models.IEEE Trans.PAMI.1983,5:25-39.
[49]杨平.CBIR中图像匹配的研究:(硕士学位论文).无锡:江南大学,2003.
[50] J. Z. Wang. Semantics-sensitive integrated matching for picture libraries and biomedical image databases, Ph.D. Dissertation, Stanford University, 2000.
[51] Yixin Chen, James Z. Wang. A region-based fuzzy feature matching approach to content-based image retrieval.IEEE. PAMI,2002,24 (9):34-36..
[52] Zhang C, Wang P. A new method of color image segmentation based on intensity and hue clustering.Proceedings of 15th International Conference on Pattern Recognition. Barcelona Spain.2000, 3:613-616.
[2]Fumera G,Roli F,A Theoretical and Experimental Analysis of Linear Combinersfor Multiple Classifier Systems.Pattern Analysis and Machine Intelligence,IEEE Transactions,2005,27(6):942-956
[3]Lienhart R,Wernicke A.Localizing and segmenting text in images and videos.IEEE Trans.Circuits and Systems for Video Technology,2002,12(4):256-268.
[4]胡小锋,周勇,叶庆泰.复杂背景彩色图像中的文字分割.上海交通大学学报,2006,141-147.
[5]叶晨洲,杨杰,宣国荣.车辆牌照字符识别.上海交通大学学报,2000,672-675.
[6]王学文,丁晓青,刘长松.基于Gabor变换的高鲁棒汉字识别新方法.电子学报,2002.9:1317-1321.
[7]胡小锋,赵辉.Visual c++/MATLAB图像处理与识别实用案例精选.人民邮电出版社,2004,194-253.
[8]Li C,Ding X Q,Wu Y S.Automitic text segmentation and text recognition for video indexing.Multimedia Systems,2000,8(1):69-81.
[9]李一舒.车牌字符识别的改进算法研究:(硕士学位论文).杭州:浙江大学,2007.
[10]张引,潘云鹤.彩色汽车图像牌照定位新方法.中国图像图形学报,2001,6(4):374-377.
[11]Louka Dlagnekov.Video-based Car Surveillance:Licenes Plate,Make,and Model Recognition.CSD Master's Thesis 2005.
[12]张宏烈.支持向量机在字符识别中的应用研究.微计算机信息,2006,11:245-247.
[13]孙忠贵,王玲.数字图像直方图均衡化的自适应校正研究.计算机时代.2004,11:19-20.
[14]Cormen,T.H.Introduction to Algorithm.机械工业出版社,2006.
[15]Chen Datong,Odobez Jean-Marc,Bourlard Herve.Text detection and recognition in images and video frames.Pattern Recognition.2004,37(3):595-608.
[16]王建平,钱自拓.基于数学形态学的图像汉字笔划细化和提取.合肥工业大学学报,2005,11(28):1431-1435.
[17]李明.复杂背景下的文本定位与提取研究:(硕士学位论文).合肥:合肥工业大学,2004.
[18]许剑锋.数字视频中的文本分割的研究:(博士学位论文).广州:华南理工大学,2005.
[19]王勇,郑辉,胡德文.图像和视频中的文字获取技术.中国图像图形学报,2004,5(9):532-537.
[20]Rafael C G,Richard E W.Digital Image Processing Second Edition.BeiJing,Publish House of Electronics Industry,2004,23:150-204.
[21]Lee C N,Wu B.Chinese-character-stroke-extraction algorithm based on contour information.Pattern Recognition,1998,31(6):651-663.
[22]刘峡壁,贾云得.汉字笔段形成规律及其提取方法.计算机学报,2004,27(3):390-395.
[23]孙星明,杨茂江,刘国华.完全基于结构知识的汉字笔画抽取方法.计算机研究与发展,2000,37(5):543-550.
[24]章毓晋.图像分割.北京:科学出版社,2001.
[25]崔屹.图像处理与分析——数学形态学方法及应用.北京:科学出版社,2002,5:23-34.
[26]Yi Lu,B.Harmon,J.Trenkle and R.Vogt,"An accurate and efficient system for segmenting machine-printed text" U.S Postal Service 5th Advanced Technology Conference,WashingtonD.C.November,1992,3:93-105.
[27]石洪贵.基于机器视觉的字符识别技术研究:(硕士学位论文).合肥:合肥工业大学,2006.
[28]刘倩.基于彩色图像的文本区域提取研究:(硕士学位论文).成都:西南交通大学,2003.
[29]章东平.视频文本的提取:(博士学位论文).杭州:浙江大学,2006.
[30]王学文,丁晓青,刘长松.基于Gabor变换的高鲁棒汉字识别新方法,电子学报,2002,9:1317-1322.
[31]陈芳信,马翔,姜新等.基于多神经网络的手写数字综合识别系统.信息技术,2003,27(9):1-4.
[32]孙亦南,刘伟军,王越超.基于儿何不变量的图像特征识别.计算机应用与软件,2004,12,21(12):1-3.
[33]孙即祥.模式识别中的特征提取与计算机视觉不变量.北京:国防工业出版社,2001.
[34]Zhe Chen.Extracting Projective Invariant of 3D Line Set from a SingleView.IEEE International Conference on Image Processing,2004,33:2119-2122.
[35]Sugimoto A.Geometric invariant of noncoplanar lines in a single view.Proceedings of International Conference on Pattern Recognition,Jerusalem,Israel 1994,5:190-195.
[36]Senem Velipasalar,Wayne Wolf.Recovering Field of View Lines by Using Projective Invariants.IEEE International Conference on Image Processing,2004,32:3069-3072.
[37]夏良正.数字图像处理.东南大学出版社,1999.
[38]容观澳.计算机图像处理.清华大学出版社,2000.
[39]R.M.Haralick,K.Shanmugam,I.Dinstein.Texture Feature for Image Classification,IEEE Trans.Systems Man Cybernet.1973,3(1):610-621.
[40]M.M.Galloway.Texture Classification Using Gray-level Run lengths,CLIP 4,1975,44:172-179.
[41]苏厚胜.车牌识别系统的设计与实现.可编程控制与工厂自动化,2005,3(3):78-90.
[42]孟国强.车辆车牌识别系统的原理及算法研究.计算机应用,2001,15(4):548-570.
[43]冯桂.图像直方图不变特性在影像匹配定位中的应用.计算机辅助设计与图形学学报,2000,12(2):144-200.
[44]蒋伟峰,刘济林.基于改进型CLAFIC学习子空间算法的有限汉字集识别.软件学报,2000,7:679-684.
[45]阎建国,高亮,卢京潮.图象处理技术在车牌识别中的应用.计算机应用,2000,22(1):17-18.
[46]楼兵军.基于区域的图像匹配算法的关键技术研究:(硕士学位论文).西安:西安电子科技大学,2006.
[47]孟丽娜,田玉敏.基于区域的图像检索技术研究:(硕士学位论文).西安:西安电子科技大学,2005.
[48]G.R.Cross,A.K.Jain,Markov random field texture models.IEEE Trans.PAMI.1983,5:25-39.
[49]杨平.CBIR中图像匹配的研究:(硕士学位论文).无锡:江南大学,2003.
[50] J. Z. Wang. Semantics-sensitive integrated matching for picture libraries and biomedical image databases, Ph.D. Dissertation, Stanford University, 2000.
[51] Yixin Chen, James Z. Wang. A region-based fuzzy feature matching approach to content-based image retrieval.IEEE. PAMI,2002,24 (9):34-36..
[52] Zhang C, Wang P. A new method of color image segmentation based on intensity and hue clustering.Proceedings of 15th International Conference on Pattern Recognition. Barcelona Spain.2000, 3:613-616.
© 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 |