机刻字符自动识别技术的研究
|
摘要
在工业生产中,企业和生产商为了对生产的产品(如发动机、机械产品和电子产品等)进行更好的生产管理、质量控制和产品跟踪调查,采用机械加工的方式直接给产品刻上字母标号或者数字编码作为一种重要的鉴别手段。为尽量消除人工识别产品编号或数字编码效率低、易出错的弊端,研究机刻数字自动识别技术对于提高企业的自动化水平具有重要的现实意义和实用价值。
本课题研究的对象是刻写在枪械上的数字字符,数字字符是与背景同色的立体字符,而目前广泛研究的文字字符识别、邮政编码识别和票据数字识别都是二维平面字符。正是因为这个特点,很多现有的字符识别研究成果难以直接应用在机刻字符识别,需要研究与课题对象相适应的字符识别方法。论文的主要研究内容主要有以下方面:
介绍了机械工件数字自动识别系统的组成部分。为提高采集的图像质量,提高工件字符识别的整体性能,根据系统结构、硬件性能参数和图像自身特点选取适合的硬件。实验研究发现在识别过程中合适的照明方式和光源是系统成败的关键,并进一步研究了不同的光源照明方式对拍摄图像质量的影响,成功研制出适合系统需要的LED照明光源。
研究了工件字符预处理的方法。由于工业现场环境恶劣、光照不均匀和枪械工件本身有油渍和划痕等不利因素导致拍摄的图像降质、模糊、噪声污染严重。针对图像降质退化导致后期图像难以分割和识别率低的问题,结合机械工件自身的图像特点,分析研究了灰度均衡、图像增强和图像分割等图像处理的关键技术。通过C语言在VisualC++平台上编程实现算法,分析验证了不同算法的优缺点,提出适合工件字符的预处理方法。
在预处理的基础上研究了工件字符的分割提取方法。针对动态阈值和全局阈值各自的优点,结合工件字符图像的特点,提出了基于LOG算子的局部动态阈值和最大类间方差的全局阈值相结合的新方法。新算法已应用于实际系统,取得了较好的实验结果。利用搭建的系统实验平台,对采集的大量枪械工件字符图像进行识别验证实验。进一步测试优化软件系统,提高系统的稳定性和高效性。
In industrial production, enterprises and manufacturers take mechanical processing ways to engrave letter directly to the product label or digital code as an important means of identification,in order to make better production management, quality control and tracking surveys for the products (such as engines, machinery and electronic products, etc.) . To avoid defects of manual detection and identification product number or digital code inefficient and error-prone, researching engraved digital automatic identification technology to enhance the level of automation of enterprises which has important practical significance and value.
The object of this study is digital code engraved in the firearms, numeric characters with the same color as the background are solid characters. At present, a wide research of current text character recognition, postal code recognition and note identification are two-dimensional planar characters. For this feature ,a lot of existing research results of character recognition are difficult to be applied directly to machine engraved characters, which needs to study identification ways to adapt the object of the study .The researches of this thesis includes following parts:
Introduce the parts of automatic identification systems of mechanism workpiece number. To improve the quality of the collected images, enhancing the overall performance of character recognition, the appropriate hardware is selected appropriately according to the system structure, hardware performance parameters and the inherent characteristics of the images. Experimental study found that: in the recognition process, the appropriate methods and a good light source lighting system is the key to success .Studying the different lighting ways affect on the quality of recorded images, successfully Fabricated the LED lighting system, which matches to the system needs.
The Study of the workpiece character pretreatment methods. As unfavorable factors, for instance poor environment of the industrial site, uneven illumination, and the oil stain and scratches on the firearms, result in the quality of images degraded shot, fuzzy and noise pollution seriously. The problem of quality degradation caused the image difficult to separate later and the low rate of identification, combining the own image characteristics of mechanical device, analyzed the key image processing techniques, such as the gray balance, image enhancement and image segmentation and so on. Using the C language under the VisualC + + platform to realize algorithm, and analysed the advantages and disadvantages of different algorithms, proposed pretreatment method for character parts.
On the basis of pretreatment method, studied the segmentation and extraction of the workpiece characters. With the dynamic threshold and global threshold of their respective advantages, and the characteristics of the workpiece image character, proposed the local dynamic threshold of LOG operator and the new algorithm based on Otsu's global threshold. The new algorithm has been applied to practical systems, and achieved good results. With an experimental platform system, do identification and verification experiments on a large number of firearms collected artifacts, tested and optimized software system, which improved system stability and efficiency.
本课题研究的对象是刻写在枪械上的数字字符,数字字符是与背景同色的立体字符,而目前广泛研究的文字字符识别、邮政编码识别和票据数字识别都是二维平面字符。正是因为这个特点,很多现有的字符识别研究成果难以直接应用在机刻字符识别,需要研究与课题对象相适应的字符识别方法。论文的主要研究内容主要有以下方面:
介绍了机械工件数字自动识别系统的组成部分。为提高采集的图像质量,提高工件字符识别的整体性能,根据系统结构、硬件性能参数和图像自身特点选取适合的硬件。实验研究发现在识别过程中合适的照明方式和光源是系统成败的关键,并进一步研究了不同的光源照明方式对拍摄图像质量的影响,成功研制出适合系统需要的LED照明光源。
研究了工件字符预处理的方法。由于工业现场环境恶劣、光照不均匀和枪械工件本身有油渍和划痕等不利因素导致拍摄的图像降质、模糊、噪声污染严重。针对图像降质退化导致后期图像难以分割和识别率低的问题,结合机械工件自身的图像特点,分析研究了灰度均衡、图像增强和图像分割等图像处理的关键技术。通过C语言在VisualC++平台上编程实现算法,分析验证了不同算法的优缺点,提出适合工件字符的预处理方法。
在预处理的基础上研究了工件字符的分割提取方法。针对动态阈值和全局阈值各自的优点,结合工件字符图像的特点,提出了基于LOG算子的局部动态阈值和最大类间方差的全局阈值相结合的新方法。新算法已应用于实际系统,取得了较好的实验结果。利用搭建的系统实验平台,对采集的大量枪械工件字符图像进行识别验证实验。进一步测试优化软件系统,提高系统的稳定性和高效性。
In industrial production, enterprises and manufacturers take mechanical processing ways to engrave letter directly to the product label or digital code as an important means of identification,in order to make better production management, quality control and tracking surveys for the products (such as engines, machinery and electronic products, etc.) . To avoid defects of manual detection and identification product number or digital code inefficient and error-prone, researching engraved digital automatic identification technology to enhance the level of automation of enterprises which has important practical significance and value.
The object of this study is digital code engraved in the firearms, numeric characters with the same color as the background are solid characters. At present, a wide research of current text character recognition, postal code recognition and note identification are two-dimensional planar characters. For this feature ,a lot of existing research results of character recognition are difficult to be applied directly to machine engraved characters, which needs to study identification ways to adapt the object of the study .The researches of this thesis includes following parts:
Introduce the parts of automatic identification systems of mechanism workpiece number. To improve the quality of the collected images, enhancing the overall performance of character recognition, the appropriate hardware is selected appropriately according to the system structure, hardware performance parameters and the inherent characteristics of the images. Experimental study found that: in the recognition process, the appropriate methods and a good light source lighting system is the key to success .Studying the different lighting ways affect on the quality of recorded images, successfully Fabricated the LED lighting system, which matches to the system needs.
The Study of the workpiece character pretreatment methods. As unfavorable factors, for instance poor environment of the industrial site, uneven illumination, and the oil stain and scratches on the firearms, result in the quality of images degraded shot, fuzzy and noise pollution seriously. The problem of quality degradation caused the image difficult to separate later and the low rate of identification, combining the own image characteristics of mechanical device, analyzed the key image processing techniques, such as the gray balance, image enhancement and image segmentation and so on. Using the C language under the VisualC + + platform to realize algorithm, and analysed the advantages and disadvantages of different algorithms, proposed pretreatment method for character parts.
On the basis of pretreatment method, studied the segmentation and extraction of the workpiece characters. With the dynamic threshold and global threshold of their respective advantages, and the characteristics of the workpiece image character, proposed the local dynamic threshold of LOG operator and the new algorithm based on Otsu's global threshold. The new algorithm has been applied to practical systems, and achieved good results. With an experimental platform system, do identification and verification experiments on a large number of firearms collected artifacts, tested and optimized software system, which improved system stability and efficiency.
引文
[1]张洪刚,陈光,郭军编.图像处理与识别[M].北京:北京邮电大学出版社, 2006.
[2]龚声荣,?刘纯平,?王强等.数字图像处理与分析[M].北京:清华大学出版社,?2006.
[3]章毓晋.图像处理与分析[M].北京:清华大学出版社, 2006.
[4]李俊山,李旭辉.数字图像处理[M].北京:清华大学出版社, 2007.
[5]唐良瑞,马全明,景晓军.图像处理实用技术[M].北京:化学工业出版社, 2002.
[6] Rafael C.Gonzalez, Richard E.Woods著,阮秋琦,阮宇智译.数字图像处理[M].北京:电子工业出版社, 2008.
[7]杨枝灵,王开. Visual C++数字图像获取处理及实践应用[M].北京:人民邮电出版社, 2003.
[8]卢瑞文.自动识别技术[M].北京:化学工业出版社, 2005.
[9]李建美.标牌压印字符图像获取与处理中的关键技术研究[D].山东大学, 2008.
[10]沈庭芝,方子义.数字图像处理及模式识别[M].北京:北京理工人学出版社, 2000.
[11]夏良正,李久贤.数字图像处理[M].南京:东南大学出版社, 2005.
[12]杨淑英.图像模式识别—VC++技术实现[M].北京:清华大学出版社, 2005, 7.
[13] J.P.M arques de sa著,吴逸飞译.模式识别—原理、方法及应用[M].北京:清华大学出版社, 2002.
[14]求是科技,张宏林. Visual C++数字图像模式识别技术及工程实践[M].北京:人民邮电出版社, 2003, 2.
[15]吴捷,陈德智,郭成志. Sobel边缘检测算法的变异实现图像增强[J].激光与红外, 2008, 38( 6).
[16]李恩玉,杨平先.基于图像边缘增强组合算法的研究[J].红外技术, 2009, 2(2).
[17]乔闹生,叶玉堂,黄永林,刘霖. PCB显微图像缺陷圆孔检测方法研究[J].光电子·激光2009, 20(7).
[18]于殿泓.图像检测与处理技术[M]西安:西安电子科技大学出版社, 2006.
[19]王庆有. CCD应用技术[M].天津:天津大学出版社, 2000, 11.
[20]蔡文贵. CCD技术及应用[M].北京:电子工业出版社, 1992, 11.
[21]徐启明,张启衡,陈强.基于DDR模组阵列的超高速数字图像存储技术[J].光学精密工程,2009,17(1).
[22]孔祥伟.组合光源与图像处理算法在工件表面缺陷检测中的应用[D].天津大学, 2007.
[23]闫枫,吴斌.视觉检测系统中的光源照明方法[J].自动测量与控制, 2006, 25(11).
[24]吴雯岑,赵辉,刘伟文,陶卫.精密视觉测量中照明对图像质量的影响[J].上海交通大学学报, 2009, 43(6).
[25]刘明瑞.数字图像预处理的硬件设计与实现方法[D].上海:上海交通大学, 2007.
[26]彭碧霞.车牌图像自动识别系统预处理算法研究[J].武汉大学学报(工学版), 2006, 39(3):131-134.
[27]田雪冰,韩绍坤.零件钢印号码自动识别预处理算法研究[J].光学技术, 2006, 32(8):557~559.
[28]求实科技. Visual C++数字图像处理典型算法及实现[M].北京:人民邮电出版社, 2003.
[29]何斌. Visual C++数字图像处理[M].北京:人民邮电出版社, 2001.
[30]李卫华.数字图像预处理与融合方法研究[D].西安:西北工业大学, 2006.
[31]徐哲,楼文高.基于模版对比的手写体数字识别神经网络模型[J].计算机工程与应用, 2008, 44(9).
[32]孙即祥.图像处理[M].北京:科学出版社,2004 9
[33]艾海舟,武勃等译.图像处理、分析与机器视觉(第二版)[M].北京:人们邮电大学出版社, 2003, 9.
[34] Unnikrishnan R, Pantofaru C, Hebert M. Toward Objective Evaluation of Image Segmentation Algorithms [J].?Pattern Analysis and Machine Intelligence:2007, 29(6):929-944.
[35] Maya R.Gupta, Nathaniel P.Jacobson, Eric k.Garcia. OCR binarization and image pre-processing for searching historical documents[J].Pattern Recognition 40(2007):389-397.
[36] Bieniecki Wojciech, Grabowski, Szymon. Image preprocessing for improving OCR accuracy[C]. Perspective Technologies and Methods in MEMS Design, Lviv-Polyana, July 30, 2007:75-80.
[37] Yun Eun-Kyung, Cho Sung-Bae. Adaptive fingerprint image enhancement with fingerprint image quality analysis [J].Image and Vision Computing (S0262-8856),2006, 24(1):101-110.
[38]李建美,路长厚,李学勇.一种基于图像分层的标牌压印字符分割方法[J].光电子·激光, 2008, 19(6):818-822.
[39]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[J].计算机学报, 2004,,27(1):130~135.
[40]王晓健.车牌定位与字符分割算法研究及实现[D].北京邮电大学, 2009.
[41]李了了.工业现场字符识别方法的研究[D].合肥工业大学, 2003.
[42]李国平.基于莫尔技术的标牌凹凸字符图像获取与识别研究[D].山东大学, 2007.
[43]丁劲生.计算机视觉中的字符识别及软件开发[D].西北工业大学, 2005.
[44]李学勇.金属标牌压印凹凸字符的特征提取和识别方法研究[D].山东大学, 2008.
[45]文颖.数字字符识别及其应用研究[D].上海交通大学, 2009.
[46]申小阳.数字仪表动态显示字符的计算机自动识别[D].浙江工业大学, 2005.
[47]王君.数字字符检测与识别方法研究[D].华中科技大学, 2007.
[48]周小龙,张小洪,冯欣.基于视觉显著图的车牌定位算法[J].光电工程, 2009, 36(11):145-150.
[49]严国萍,戴若愚,潘晴,刘袁缘.基于LOG算子的自适应图像边缘检测方法[J].华中科技大学学报(自然科学版), 2008, 36(3):85-87.
[50] Wong W.C.K, Chung A.C.S. Bayesian image segmentation using local iso-intensity structural orientation [J]. Image Processing, 2005, 14(10):1512 - 1523.
[51] Erikson Mats. Two preprocessing techniques based on grey level and geometric thickness to improve segmentation results [J]. Pattern Recognition Letters(S0167-8655) 2006, 7(3):160-166.
[52] Chang Shyang-Lih, Chen Li-Shien, Chung Yun-Chung, Chen Sei-Wan. Automatic license plate recognition[J]. IEEE Transactions on Intelligent Transportation Systems(S1524-9050),2004,5:42-53.
[2]龚声荣,?刘纯平,?王强等.数字图像处理与分析[M].北京:清华大学出版社,?2006.
[3]章毓晋.图像处理与分析[M].北京:清华大学出版社, 2006.
[4]李俊山,李旭辉.数字图像处理[M].北京:清华大学出版社, 2007.
[5]唐良瑞,马全明,景晓军.图像处理实用技术[M].北京:化学工业出版社, 2002.
[6] Rafael C.Gonzalez, Richard E.Woods著,阮秋琦,阮宇智译.数字图像处理[M].北京:电子工业出版社, 2008.
[7]杨枝灵,王开. Visual C++数字图像获取处理及实践应用[M].北京:人民邮电出版社, 2003.
[8]卢瑞文.自动识别技术[M].北京:化学工业出版社, 2005.
[9]李建美.标牌压印字符图像获取与处理中的关键技术研究[D].山东大学, 2008.
[10]沈庭芝,方子义.数字图像处理及模式识别[M].北京:北京理工人学出版社, 2000.
[11]夏良正,李久贤.数字图像处理[M].南京:东南大学出版社, 2005.
[12]杨淑英.图像模式识别—VC++技术实现[M].北京:清华大学出版社, 2005, 7.
[13] J.P.M arques de sa著,吴逸飞译.模式识别—原理、方法及应用[M].北京:清华大学出版社, 2002.
[14]求是科技,张宏林. Visual C++数字图像模式识别技术及工程实践[M].北京:人民邮电出版社, 2003, 2.
[15]吴捷,陈德智,郭成志. Sobel边缘检测算法的变异实现图像增强[J].激光与红外, 2008, 38( 6).
[16]李恩玉,杨平先.基于图像边缘增强组合算法的研究[J].红外技术, 2009, 2(2).
[17]乔闹生,叶玉堂,黄永林,刘霖. PCB显微图像缺陷圆孔检测方法研究[J].光电子·激光2009, 20(7).
[18]于殿泓.图像检测与处理技术[M]西安:西安电子科技大学出版社, 2006.
[19]王庆有. CCD应用技术[M].天津:天津大学出版社, 2000, 11.
[20]蔡文贵. CCD技术及应用[M].北京:电子工业出版社, 1992, 11.
[21]徐启明,张启衡,陈强.基于DDR模组阵列的超高速数字图像存储技术[J].光学精密工程,2009,17(1).
[22]孔祥伟.组合光源与图像处理算法在工件表面缺陷检测中的应用[D].天津大学, 2007.
[23]闫枫,吴斌.视觉检测系统中的光源照明方法[J].自动测量与控制, 2006, 25(11).
[24]吴雯岑,赵辉,刘伟文,陶卫.精密视觉测量中照明对图像质量的影响[J].上海交通大学学报, 2009, 43(6).
[25]刘明瑞.数字图像预处理的硬件设计与实现方法[D].上海:上海交通大学, 2007.
[26]彭碧霞.车牌图像自动识别系统预处理算法研究[J].武汉大学学报(工学版), 2006, 39(3):131-134.
[27]田雪冰,韩绍坤.零件钢印号码自动识别预处理算法研究[J].光学技术, 2006, 32(8):557~559.
[28]求实科技. Visual C++数字图像处理典型算法及实现[M].北京:人民邮电出版社, 2003.
[29]何斌. Visual C++数字图像处理[M].北京:人民邮电出版社, 2001.
[30]李卫华.数字图像预处理与融合方法研究[D].西安:西北工业大学, 2006.
[31]徐哲,楼文高.基于模版对比的手写体数字识别神经网络模型[J].计算机工程与应用, 2008, 44(9).
[32]孙即祥.图像处理[M].北京:科学出版社,2004 9
[33]艾海舟,武勃等译.图像处理、分析与机器视觉(第二版)[M].北京:人们邮电大学出版社, 2003, 9.
[34] Unnikrishnan R, Pantofaru C, Hebert M. Toward Objective Evaluation of Image Segmentation Algorithms [J].?Pattern Analysis and Machine Intelligence:2007, 29(6):929-944.
[35] Maya R.Gupta, Nathaniel P.Jacobson, Eric k.Garcia. OCR binarization and image pre-processing for searching historical documents[J].Pattern Recognition 40(2007):389-397.
[36] Bieniecki Wojciech, Grabowski, Szymon. Image preprocessing for improving OCR accuracy[C]. Perspective Technologies and Methods in MEMS Design, Lviv-Polyana, July 30, 2007:75-80.
[37] Yun Eun-Kyung, Cho Sung-Bae. Adaptive fingerprint image enhancement with fingerprint image quality analysis [J].Image and Vision Computing (S0262-8856),2006, 24(1):101-110.
[38]李建美,路长厚,李学勇.一种基于图像分层的标牌压印字符分割方法[J].光电子·激光, 2008, 19(6):818-822.
[39]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[J].计算机学报, 2004,,27(1):130~135.
[40]王晓健.车牌定位与字符分割算法研究及实现[D].北京邮电大学, 2009.
[41]李了了.工业现场字符识别方法的研究[D].合肥工业大学, 2003.
[42]李国平.基于莫尔技术的标牌凹凸字符图像获取与识别研究[D].山东大学, 2007.
[43]丁劲生.计算机视觉中的字符识别及软件开发[D].西北工业大学, 2005.
[44]李学勇.金属标牌压印凹凸字符的特征提取和识别方法研究[D].山东大学, 2008.
[45]文颖.数字字符识别及其应用研究[D].上海交通大学, 2009.
[46]申小阳.数字仪表动态显示字符的计算机自动识别[D].浙江工业大学, 2005.
[47]王君.数字字符检测与识别方法研究[D].华中科技大学, 2007.
[48]周小龙,张小洪,冯欣.基于视觉显著图的车牌定位算法[J].光电工程, 2009, 36(11):145-150.
[49]严国萍,戴若愚,潘晴,刘袁缘.基于LOG算子的自适应图像边缘检测方法[J].华中科技大学学报(自然科学版), 2008, 36(3):85-87.
[50] Wong W.C.K, Chung A.C.S. Bayesian image segmentation using local iso-intensity structural orientation [J]. Image Processing, 2005, 14(10):1512 - 1523.
[51] Erikson Mats. Two preprocessing techniques based on grey level and geometric thickness to improve segmentation results [J]. Pattern Recognition Letters(S0167-8655) 2006, 7(3):160-166.
[52] Chang Shyang-Lih, Chen Li-Shien, Chung Yun-Chung, Chen Sei-Wan. Automatic license plate recognition[J]. IEEE Transactions on Intelligent Transportation Systems(S1524-9050),2004,5:42-53.