基于小波分析的车牌图像增强与字符识别研究
|
摘要
随着社会经济的发展和人们生活水平的提高,由于道路硬件设施的限制,智能交通越来越受到人们的重视。车牌识别系统是智能交通的重要组成部分,它在城市交通管理、高速公路收费、电子警察、智能停车场等领域有着非常重要的实际应用价值,正日益受到人们的重视。
本文结合我国车牌的特殊性,采用图像处理技术、车牌定位技术、字符分割及归一化技术、神经网络识别技术等对车牌图像进行处理和识别。主要完成的工作如下:
①将采集到的含有车牌的图像进行灰度化,通过灰度拉伸、直方图均衡化、中值滤波等对图像进行增强,并提出了改进的迭代阈值法对图像进行二值化处理。考虑到微光下车牌图像的含噪特点,结合小波变换的时频分析能力,利用小波变换对微光下的图像进行了去噪处理,取得了良好的去噪效果。
②通过对我国车牌特征的分析,和对常见车牌定位方法的研究,本文使用了将边缘检测和颜色特征相结合的车牌定位方法,实验结果证明该方法对正常光照下和微光下的车牌都具有很好的定位效果。然后采用斜率法对定位出的车牌进行倾斜校正。
③采用了一种基于灰度累加的车牌字符分割方法对车牌字符进行分割,证明对图像质量较差的车牌字符也能有较好的分割效果。然后用邻近插值法对分割出的字符进行归一化处理。
④采用小波矩对字符进行特征提取,并采用小波神经网络对字符进行分类识别,描述了相应的算法和程序设计步骤,并将识别结果与BP神经网络进行对比,通过实验说明了小波神经网络的优越性。
With the socio-economic development and people's living standard improvement, as the road hardware limitation, intelligent transportation system (ITS) has been paid more and more attention. License plate recognition (LPR) is an important part of ITS. It has very important practical application value in the urban traffic management, highway toll, electronic police, intelligence car park and other fields, which has been paid attention increasingly.
Combining with the license plate particularity in China, this paper applies image processing technology, license plate location technology, character segmentation and normalization techniques, and neural network technology,to carries out license plate image processing and recognition. The major completed work as follows:
①Taking the gray level transform to the collected image which contained license plate, then through the gray stretch, histogram equalization, median filtering schedules to improve the image effect, and proposing an improved iterative threshold method of the binary image processing. Taking into account of noisy image characteristics of license plate the license in low light level, combining the wavelet transform time-frequency analysis, this paper makes use of wavelet transform to de-noising the license plate image, and achieved an good de-noising effect.
②Through the analysis of license plate features, and the research on the common license plate location method, this paper applies the license plate location method combined the edge detection with color characteristics. Experiment results show that this method has a very good location effect to the license plate both in normal light and low light level. Then we use the slope method to do the tilt correction for the located license plate image.
③Using a character segmentation method based on gray additive to segment the vehicle license plate character, and the experiment result shows that this method has a better character segmentation effect to the license plate image with poor quality. Then do the normalized process to the segment characters with the neighboring interpolation method.
④Using the wavelet moment to extract the character feature, and then recognizing the characters with the wavelet neural network. In this paper, we have described the corresponding algorithm and programming steps. Compared with BP neural network, the experiment shows the wavelet neural network has a higher recognition rate.
本文结合我国车牌的特殊性,采用图像处理技术、车牌定位技术、字符分割及归一化技术、神经网络识别技术等对车牌图像进行处理和识别。主要完成的工作如下:
①将采集到的含有车牌的图像进行灰度化,通过灰度拉伸、直方图均衡化、中值滤波等对图像进行增强,并提出了改进的迭代阈值法对图像进行二值化处理。考虑到微光下车牌图像的含噪特点,结合小波变换的时频分析能力,利用小波变换对微光下的图像进行了去噪处理,取得了良好的去噪效果。
②通过对我国车牌特征的分析,和对常见车牌定位方法的研究,本文使用了将边缘检测和颜色特征相结合的车牌定位方法,实验结果证明该方法对正常光照下和微光下的车牌都具有很好的定位效果。然后采用斜率法对定位出的车牌进行倾斜校正。
③采用了一种基于灰度累加的车牌字符分割方法对车牌字符进行分割,证明对图像质量较差的车牌字符也能有较好的分割效果。然后用邻近插值法对分割出的字符进行归一化处理。
④采用小波矩对字符进行特征提取,并采用小波神经网络对字符进行分类识别,描述了相应的算法和程序设计步骤,并将识别结果与BP神经网络进行对比,通过实验说明了小波神经网络的优越性。
With the socio-economic development and people's living standard improvement, as the road hardware limitation, intelligent transportation system (ITS) has been paid more and more attention. License plate recognition (LPR) is an important part of ITS. It has very important practical application value in the urban traffic management, highway toll, electronic police, intelligence car park and other fields, which has been paid attention increasingly.
Combining with the license plate particularity in China, this paper applies image processing technology, license plate location technology, character segmentation and normalization techniques, and neural network technology,to carries out license plate image processing and recognition. The major completed work as follows:
①Taking the gray level transform to the collected image which contained license plate, then through the gray stretch, histogram equalization, median filtering schedules to improve the image effect, and proposing an improved iterative threshold method of the binary image processing. Taking into account of noisy image characteristics of license plate the license in low light level, combining the wavelet transform time-frequency analysis, this paper makes use of wavelet transform to de-noising the license plate image, and achieved an good de-noising effect.
②Through the analysis of license plate features, and the research on the common license plate location method, this paper applies the license plate location method combined the edge detection with color characteristics. Experiment results show that this method has a very good location effect to the license plate both in normal light and low light level. Then we use the slope method to do the tilt correction for the located license plate image.
③Using a character segmentation method based on gray additive to segment the vehicle license plate character, and the experiment result shows that this method has a better character segmentation effect to the license plate image with poor quality. Then do the normalized process to the segment characters with the neighboring interpolation method.
④Using the wavelet moment to extract the character feature, and then recognizing the characters with the wavelet neural network. In this paper, we have described the corresponding algorithm and programming steps. Compared with BP neural network, the experiment shows the wavelet neural network has a higher recognition rate.
引文
[1]陈艳,何春明.智能交通系统应用现状及其存在问题分析[J].交通标准化, 2007, (8).
[2] Rafael C. Gonzales Richard E. Woods等著,阮秋琦,阮宇智等译[M].数字图像处理(第二版)电子工业出版社, 2003,3.
[3]高成. Matlab图像处理与应用[M].国防工业出版社, 2007,4:132-133.
[4]路小波,张光华.基于二值图像的车牌精确定位方法[J].东南大学学报(自然科学版), 2005, 35(6):972-974.
[5] Bernsen J. Dynamic thresholding of gray-level images[A]. Proc.8th Int. Conf.on Pattern Recognition[C]. LOSAngeles: IEEE Computer Society Press, 1986:1251-1255.
[6] Nostu. A Thrshold Selection Method form Gray-level Histograms. IEEE Trans Svst, Man Cvbernet. 1979, SMC-9:62-66.
[7]张保民.成像系统分析导论(第1版)[M].北京:国防工业出版社, 1992.
[8]高珊,刘万春,朱玉文.基于SVM的车牌字符分割和识别方法[J].微电子学与计算机, 2005, 22(6):34-36.
[9]陈寅鹏,丁晓清.复杂车辆图像中的车牌定位与字符分割方法[J].红外与激光工程, 2004, 33(1):29-33.
[10]付仲良,黄书强,陈氓,陈江平.货车图像车牌区快速定位及字符切割算法[J].计算机工程与设计, 2003, 24(1):77-79.
[11]朱风云,曹晓光等.机动车车牌自动识别系统与VMLA定位算法[J].中国图像图形学报, 2003, 8(6):679-682,.
[12] Taleb-Ahmed, A.Hamad, D.Tilmant. Vehicle license plate recognition in marketing application Intelligent Vehicles Symposium. IEEE, 9-11 June 2003:90-94.
[13]韩智广等.车牌分割与矫正[J].计算机工程与应用, 2003, (9):210-212.
[14]蔡钦涛,方水良,任俊.基于边缘生长的车牌定位新方法[J].公路交通科技, 2004,21(11):110-113.
[15]李文举,梁德群,张旗,樊鑫.基于边缘颜色对的车牌定位新方法[J].计算机学报, 2004, 27(2):204-208.
[16] Xifan Shi, Weizhong Zhao, and Yonghang Shen.. Automatic license plate recognition system based on color image processing[C].ICCSA 2005, LNCS 3483, 1159-1168, 2005.
[17]高朝晖,黄卫.基于彩色图像车牌分割研究[J].公路交通科技, 2004, 21(8):114-117.
[18]许剑峰,黎绍发.车牌识别中的颜色分析[J].计算机工程与应用, 2004, 40(25):230-232.
[19]杨枝灵,王开等. Visual C++数字图像获取处理及实践应用(第一版)[M].北京:人民邮电出版社, 2003:174-175.
[20]樊孝宏,戚飞虎.一种基于纹理和颜色综合特征的车牌定位新方法[J].计算机工程, 2004,30(13):125-127.
[21]张引,潘云鹤.彩色汽车图像牌照定位新方法[J].中国图像图形学报, 2001, 6(4):374～377
[22]周铁平,王庆.一种新的汽车牌照快速定位方法[J].微计算机应用, 2007, 28(4):343~348
[23]吴传孙,汽车牌照自动识别技术研究[D].硕士学位论文.江西师范大学, 2003.
[24]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, (1): 130-135.
[25]杨立刚,张兴会,李兰友.车牌字符倾斜校正方法的研究[J].仪器仪表学报, 2004, 25(4刊):696-697.
[26]叶青,张春华.汽车牌照实时几何畸变校正方法[J].红外与激光工程, 2004(3), vol 3 No.2.
[27]吴进军,杜树新.车牌字符分割新方法[J].工业控制计算机, 2005, 8(4):69-75.
[28] Yanamura Y, Goto M.,Nishiyama D, Soga M, Nakatani H, Saji H. Extraction and tracking of the license plate using Hough transform and voted block matching. IEEE IV2003 Intelligent Vehicles Symposium. 2003:243-246.
[29]王雨晨.基于图像边缘处理和模糊识别的车牌识别方法[D].硕士学位论文.东北大学, 2003.
[30] Sirithinaphong, T; Chamnongthai, K., Extraction of car license plate using motor vehicle regulation and character pattern recognition, Circuits and Systems, 1998. IEEE APCCAS 1998.
[31] Da Rocha Gesualdi, A.; Manoel de Seixas, J. Character recognition in car license plate based on principal components and neural processing, Neural Network, 2002. SBRN 2002.
[32]孙虎儿.基于神经网络的优化设计及应用[J].北京:国防工业出版社, 2009.5:30-32.
[33]田雨波.混合神经网络技术[M].北京:科学出版社, 2009:297-318
[34] Merz C J, Pazzani M J. A principal components approach to combining regression estinates. Machine Learning, 1999, 36(1,2):9-32.
[35]万宜,基于小波神经网络的车牌自动识别研究[D],东南大学硕士学位论文, 2004.3:13-16.
[36]夏劲,郭红卫.国内外城市智能交通系统的发展概况与趋势及启示[J].科技进步与对策,2003, 1:176-179.
[37]朱虹等编著.数字图像处理基础[M].北京:科学出版社,2005.
[38]陶军.车牌识别技术研究与实现[D].南京:南京理工大学模式识别与智能系统, 2004.
[39] Cheoman Wu, Lei Chan On, Chan Hon Weng et al. Amacao License Plate Recognition System . In: Proceeding of the Fourth International Conference on Machine Learning and Cybernetics, Guangzhou,2005:4506-4510.
[40]邱刚,王养丽.基于边缘特征和神经网络的汽车牌照定位方法[J].微机发展, 2005.4:30-32.
[41] AN Yong, ZHANG Gao-wei. A Gray-Image-Based License Plate Recognition System[J]. Computer Engineering & Science, 2006,28(2):61-65.
[42] Danian Zheng,Yannan Zhao,Jiaxin Wang. An efficient method of license plate location[J]. Pattern Recongnition Letters.2005,26(15):2431-2438.
[43] Sunghoon Kim, Daechul Kim, Younbok Ryu, and Gyeonghwan Kim. A robust license-plate-extraction method under complex image conditions[C]. 16th international conference on pattern recognition,2002:176-179.
[44]沈世宴,盛祤智.基于边缘检测的车牌图像分割技术[J].自动化技术与应用, 2004,23(3):24-26.
[45] Bai Hongliang , Liu Changping. A hybrid license plate extraction method based on edge statistics and morphology [C]. ICPR.2004:831-834.
[46] David Llorens,Andress Marzal, Vicente Palazbn, and Juan M.Vilar. Car license plate extraction and recognition based on connected components analysis and HMM Decoding [C].IbPRIA 2005:571-578.
[47] Regis C.P. Marques, Fotima N.S. Medeiros, Jilseph Lopes Silva, and Cassius M.Laprano. License vehicle plates localization using Maximum Correlation [C]. SSPR&SPR 2004, LNCS 3138:470-476.
[48]金一粟,袁宝民,于万波,魏小鹏.基于分形盒子维数的车牌定位方法[J].计算机应用研究, 2002, 9:40-44.
[49]卢雅琴,邬凌超.基于数学形态学的车牌定位方法[J].计算机工程, 2005,31(3):224-226.
[50] Suryanarayana P V,Suman K Mitra, Asim Banerjee, et al. A Morphology Based Approach for Car License Plate Extraction [C]. Chennai, IndiaL, IEEE,2005.
[51] LI G, ZENG R L, Lin L. Research on vehicle license plate location based on neural networks C.Proceedings of IEEE First International Conference on Innovative Computing, Information and Control, 2006:174-177.
[2] Rafael C. Gonzales Richard E. Woods等著,阮秋琦,阮宇智等译[M].数字图像处理(第二版)电子工业出版社, 2003,3.
[3]高成. Matlab图像处理与应用[M].国防工业出版社, 2007,4:132-133.
[4]路小波,张光华.基于二值图像的车牌精确定位方法[J].东南大学学报(自然科学版), 2005, 35(6):972-974.
[5] Bernsen J. Dynamic thresholding of gray-level images[A]. Proc.8th Int. Conf.on Pattern Recognition[C]. LOSAngeles: IEEE Computer Society Press, 1986:1251-1255.
[6] Nostu. A Thrshold Selection Method form Gray-level Histograms. IEEE Trans Svst, Man Cvbernet. 1979, SMC-9:62-66.
[7]张保民.成像系统分析导论(第1版)[M].北京:国防工业出版社, 1992.
[8]高珊,刘万春,朱玉文.基于SVM的车牌字符分割和识别方法[J].微电子学与计算机, 2005, 22(6):34-36.
[9]陈寅鹏,丁晓清.复杂车辆图像中的车牌定位与字符分割方法[J].红外与激光工程, 2004, 33(1):29-33.
[10]付仲良,黄书强,陈氓,陈江平.货车图像车牌区快速定位及字符切割算法[J].计算机工程与设计, 2003, 24(1):77-79.
[11]朱风云,曹晓光等.机动车车牌自动识别系统与VMLA定位算法[J].中国图像图形学报, 2003, 8(6):679-682,.
[12] Taleb-Ahmed, A.Hamad, D.Tilmant. Vehicle license plate recognition in marketing application Intelligent Vehicles Symposium. IEEE, 9-11 June 2003:90-94.
[13]韩智广等.车牌分割与矫正[J].计算机工程与应用, 2003, (9):210-212.
[14]蔡钦涛,方水良,任俊.基于边缘生长的车牌定位新方法[J].公路交通科技, 2004,21(11):110-113.
[15]李文举,梁德群,张旗,樊鑫.基于边缘颜色对的车牌定位新方法[J].计算机学报, 2004, 27(2):204-208.
[16] Xifan Shi, Weizhong Zhao, and Yonghang Shen.. Automatic license plate recognition system based on color image processing[C].ICCSA 2005, LNCS 3483, 1159-1168, 2005.
[17]高朝晖,黄卫.基于彩色图像车牌分割研究[J].公路交通科技, 2004, 21(8):114-117.
[18]许剑峰,黎绍发.车牌识别中的颜色分析[J].计算机工程与应用, 2004, 40(25):230-232.
[19]杨枝灵,王开等. Visual C++数字图像获取处理及实践应用(第一版)[M].北京:人民邮电出版社, 2003:174-175.
[20]樊孝宏,戚飞虎.一种基于纹理和颜色综合特征的车牌定位新方法[J].计算机工程, 2004,30(13):125-127.
[21]张引,潘云鹤.彩色汽车图像牌照定位新方法[J].中国图像图形学报, 2001, 6(4):374～377
[22]周铁平,王庆.一种新的汽车牌照快速定位方法[J].微计算机应用, 2007, 28(4):343~348
[23]吴传孙,汽车牌照自动识别技术研究[D].硕士学位论文.江西师范大学, 2003.
[24]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, (1): 130-135.
[25]杨立刚,张兴会,李兰友.车牌字符倾斜校正方法的研究[J].仪器仪表学报, 2004, 25(4刊):696-697.
[26]叶青,张春华.汽车牌照实时几何畸变校正方法[J].红外与激光工程, 2004(3), vol 3 No.2.
[27]吴进军,杜树新.车牌字符分割新方法[J].工业控制计算机, 2005, 8(4):69-75.
[28] Yanamura Y, Goto M.,Nishiyama D, Soga M, Nakatani H, Saji H. Extraction and tracking of the license plate using Hough transform and voted block matching. IEEE IV2003 Intelligent Vehicles Symposium. 2003:243-246.
[29]王雨晨.基于图像边缘处理和模糊识别的车牌识别方法[D].硕士学位论文.东北大学, 2003.
[30] Sirithinaphong, T; Chamnongthai, K., Extraction of car license plate using motor vehicle regulation and character pattern recognition, Circuits and Systems, 1998. IEEE APCCAS 1998.
[31] Da Rocha Gesualdi, A.; Manoel de Seixas, J. Character recognition in car license plate based on principal components and neural processing, Neural Network, 2002. SBRN 2002.
[32]孙虎儿.基于神经网络的优化设计及应用[J].北京:国防工业出版社, 2009.5:30-32.
[33]田雨波.混合神经网络技术[M].北京:科学出版社, 2009:297-318
[34] Merz C J, Pazzani M J. A principal components approach to combining regression estinates. Machine Learning, 1999, 36(1,2):9-32.
[35]万宜,基于小波神经网络的车牌自动识别研究[D],东南大学硕士学位论文, 2004.3:13-16.
[36]夏劲,郭红卫.国内外城市智能交通系统的发展概况与趋势及启示[J].科技进步与对策,2003, 1:176-179.
[37]朱虹等编著.数字图像处理基础[M].北京:科学出版社,2005.
[38]陶军.车牌识别技术研究与实现[D].南京:南京理工大学模式识别与智能系统, 2004.
[39] Cheoman Wu, Lei Chan On, Chan Hon Weng et al. Amacao License Plate Recognition System . In: Proceeding of the Fourth International Conference on Machine Learning and Cybernetics, Guangzhou,2005:4506-4510.
[40]邱刚,王养丽.基于边缘特征和神经网络的汽车牌照定位方法[J].微机发展, 2005.4:30-32.
[41] AN Yong, ZHANG Gao-wei. A Gray-Image-Based License Plate Recognition System[J]. Computer Engineering & Science, 2006,28(2):61-65.
[42] Danian Zheng,Yannan Zhao,Jiaxin Wang. An efficient method of license plate location[J]. Pattern Recongnition Letters.2005,26(15):2431-2438.
[43] Sunghoon Kim, Daechul Kim, Younbok Ryu, and Gyeonghwan Kim. A robust license-plate-extraction method under complex image conditions[C]. 16th international conference on pattern recognition,2002:176-179.
[44]沈世宴,盛祤智.基于边缘检测的车牌图像分割技术[J].自动化技术与应用, 2004,23(3):24-26.
[45] Bai Hongliang , Liu Changping. A hybrid license plate extraction method based on edge statistics and morphology [C]. ICPR.2004:831-834.
[46] David Llorens,Andress Marzal, Vicente Palazbn, and Juan M.Vilar. Car license plate extraction and recognition based on connected components analysis and HMM Decoding [C].IbPRIA 2005:571-578.
[47] Regis C.P. Marques, Fotima N.S. Medeiros, Jilseph Lopes Silva, and Cassius M.Laprano. License vehicle plates localization using Maximum Correlation [C]. SSPR&SPR 2004, LNCS 3138:470-476.
[48]金一粟,袁宝民,于万波,魏小鹏.基于分形盒子维数的车牌定位方法[J].计算机应用研究, 2002, 9:40-44.
[49]卢雅琴,邬凌超.基于数学形态学的车牌定位方法[J].计算机工程, 2005,31(3):224-226.
[50] Suryanarayana P V,Suman K Mitra, Asim Banerjee, et al. A Morphology Based Approach for Car License Plate Extraction [C]. Chennai, IndiaL, IEEE,2005.
[51] LI G, ZENG R L, Lin L. Research on vehicle license plate location based on neural networks C.Proceedings of IEEE First International Conference on Innovative Computing, Information and Control, 2006:174-177.