车牌定位算法的研究
|
摘要
车辆牌照识别系统(LPR)作为一个专用的计算机视觉系统,能够自动地摄取车辆图像并识别出车牌。LPR系统的研究涉及数字图像处理、计算机视觉、模式识别与人工智能等多个技术领域,其关键技术包括车牌定位、字符分割、字符识别等。其中车牌定位是指将车牌区域从车辆图像中定位并分割出来,为后续操作提供了输入信息,是整个车牌识别系统的基础。
本文在详细研究了国内外各种具有代表性的车牌定位方法基础上,结合中国车牌的特点提出了三种不同的车牌定位方法:
①.《一种基于Top-hat变换的多车牌定位改进方法》:此方法以数学形态学为基础,用Top-hat变换来提取车牌信息,再通过平滑、二值化、垂直信息提取、连通、密度计算等多个操作完成车牌定位。
②.《基于二进小波变换的多车牌定位算法》:以图像二进小波变换后的垂直方向高频子带为基础,通过后续去伪定位操作来精确定位车牌,实验表明此方法定位效果要远远好于传统的小波定位方法。
③.《基于压缩域的多车牌定位技术》:该方法以JPEG格式的车牌图像中的DCT系数为研究对象。根据DCT变换的特性,从每一个8×8模块的DCT系数选取了6个代表此模块的纹理密度用以描述该模块的垂直纹理和对角纹理。再结合车牌自身纹理的特征来定位车牌。
这三种方法各有优劣,适合于不同要求的车牌定位系统中。基于Top-hat变换的车牌定位实现简单,定位效果较好,适用范围广,通用性强,适用于实时的交通系统中。基于二进小波的车牌定位方法效果最好,但运算较复杂,耗时较多,适用于后台运作的交通系统中。基于压缩域的车牌定位方法,从压缩域出发,利用JPEG文件中DCT系数来获取交通图像中的车牌纹理特征,缩短了定位时间,且定位效果优于传统的定位算法。
License plate recognition(LPR) is a special computer vision system, which can recognize the license plate automatically from the video。The study of LPR is related to many kinds of technologies such as digital image processing, computer vision, pattern recognition and artificial intelligence. Its key techniques include license plate location, character segmentation and recognition. License plate location, which is to locate and segment the License Plate from a vehicle image, providing the inputting information for the next processes, is the basis of LPR system.
Based on the detail analysis of the typical LPR system over the world and the characteristics of Chinese license plate, three different algorithms of license plate location are presented in this paper:
①.< The application of Top-hat Transform in Multi-license Plate Localization >: Combining with the texture feature of license plates and the characteristics of Top-hat transform, the Top-hat coefficients are retained. Then the license plate can be located through smoothing, vertical direction information, four connected region detection, and density calculation and so on. The proposed algorithm was suitable for the single-license and multi-license plate detection in various illumination conditions with high accuracy.
②.:In this chapter a new algorithm which can improve the accuracy of vehicle license plate location using dyadic wavelet coefficients in vertical direction is presented. The performance of the proposed algorithm has been tested on monitoring images obtained from the practical application. Experimental results show that our algorithm has better performance than Mallat’s on location precision.
③.: A new algorithm for license plate localization using the discrete cosines transform (DCT) coefficients is proposed in this section. According to the characteristic of DCT, six AC coefficients of an 8*8 block were selected to represent the vertical and diagonal textures of a block. Then following the next processes, the license plate would have been located at last.
These three algorithms having their own advantages were suitable for the different LPR systems respectively. The algorithm based on Top-hat was easily realized and suitable for the real time LPR system. The algorithms based on dyadic wavelet transform performed best, but was complicated and time-consuming. So it was suitable for the background system. The algorithm based compressed domain using DCT coefficients in JPEG files to capture the license plate enhanced localization-efficiency rapidly. Although its effect was not as good as the two algorithms mentioned before, it was better than the traditional methods.
本文在详细研究了国内外各种具有代表性的车牌定位方法基础上,结合中国车牌的特点提出了三种不同的车牌定位方法:
①.《一种基于Top-hat变换的多车牌定位改进方法》:此方法以数学形态学为基础,用Top-hat变换来提取车牌信息,再通过平滑、二值化、垂直信息提取、连通、密度计算等多个操作完成车牌定位。
②.《基于二进小波变换的多车牌定位算法》:以图像二进小波变换后的垂直方向高频子带为基础,通过后续去伪定位操作来精确定位车牌,实验表明此方法定位效果要远远好于传统的小波定位方法。
③.《基于压缩域的多车牌定位技术》:该方法以JPEG格式的车牌图像中的DCT系数为研究对象。根据DCT变换的特性,从每一个8×8模块的DCT系数选取了6个代表此模块的纹理密度用以描述该模块的垂直纹理和对角纹理。再结合车牌自身纹理的特征来定位车牌。
这三种方法各有优劣,适合于不同要求的车牌定位系统中。基于Top-hat变换的车牌定位实现简单,定位效果较好,适用范围广,通用性强,适用于实时的交通系统中。基于二进小波的车牌定位方法效果最好,但运算较复杂,耗时较多,适用于后台运作的交通系统中。基于压缩域的车牌定位方法,从压缩域出发,利用JPEG文件中DCT系数来获取交通图像中的车牌纹理特征,缩短了定位时间,且定位效果优于传统的定位算法。
License plate recognition(LPR) is a special computer vision system, which can recognize the license plate automatically from the video。The study of LPR is related to many kinds of technologies such as digital image processing, computer vision, pattern recognition and artificial intelligence. Its key techniques include license plate location, character segmentation and recognition. License plate location, which is to locate and segment the License Plate from a vehicle image, providing the inputting information for the next processes, is the basis of LPR system.
Based on the detail analysis of the typical LPR system over the world and the characteristics of Chinese license plate, three different algorithms of license plate location are presented in this paper:
①.< The application of Top-hat Transform in Multi-license Plate Localization >: Combining with the texture feature of license plates and the characteristics of Top-hat transform, the Top-hat coefficients are retained. Then the license plate can be located through smoothing, vertical direction information, four connected region detection, and density calculation and so on. The proposed algorithm was suitable for the single-license and multi-license plate detection in various illumination conditions with high accuracy.
②.
③.
These three algorithms having their own advantages were suitable for the different LPR systems respectively. The algorithm based on Top-hat was easily realized and suitable for the real time LPR system. The algorithms based on dyadic wavelet transform performed best, but was complicated and time-consuming. So it was suitable for the background system. The algorithm based compressed domain using DCT coefficients in JPEG files to capture the license plate enhanced localization-efficiency rapidly. Although its effect was not as good as the two algorithms mentioned before, it was better than the traditional methods.
引文
[1] Parisi R, Di Claudio, Orlandi G, Car Plate Recognition by Neural Networks and Image Processing[C],Proc IEEE International Symposium on Circuits and System, USA, 1998: 195-198.
[2] Barroso,j, Dagless,E.L, Rafael,A, Bulas-Cruz,j. Number plate reading using computer vision Industrial Electronics[C],Proc IEEE International Symposium on Industrial Electronics(ISIE),Portugal, 1997:761-766 .
[3] Charl Coetzee. Based Number Plate Recognition Systems[C],Proc IEEE International Symposium on Industrial Electronics,1998:605～610.
[4]张引,潘云鹤.彩色图象定位新方法[J],中国图象图形学报,2001,6(4):374-377.
[5]郭捷,施鹏飞.基于颜色和纹理分析的车牌定位方法[J],中国图象图形学报,2002, 7(5):472-475.
[6]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,2004,12(02):204-208.
[7]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):224-225.
[8]马永力,夏秋华.基于数学形态学的车牌精定位算法的研究[J],微计算机信息,2008,24(01):227-229
[9]王锋,彭国华,赵强.一种快速、高效的车牌定位方法[J].计算机工程与应用,2007,43(11):120-123.
[10]张向东,马月涛,杨奇泽.一种基于灰度图像的车牌快速定位算法[J].电子科技,2007,212(05):173-177.
[11]赵涛,杨晓莉,王绪本,张娜.一种用于车牌定位的改进BP神经网络方法[J],算机仿真,2007,24(02):240-243.
[12]陈建坤,范春年.一种基于神经网络的车牌定位方法[J].辽宁工程技术大学学报, 2005,24(01):97-100.
[13]赵先军.基于神经网络的车牌识别技术研究[D].西安电子科技大学硕士论文,2005.
[14]杨晓敏.车牌识别关键技术研究[D],四川大学博士论文,2008.
[15]陈兆学,施鹏飞.基于灰度图像的车牌快速定位和分割方法[J],计算机工程.2006,32(09):173-177.
[16]韩永强,李世祥.基于线处理的子图像定位算法[J],计算机与现代化.1998(2):10-12.
[17]魏武,黄心汉,王敏.一种基于垂直字符边界特征的车牌定位方法[J],中国公路学报.2000,13(04):88-99.
[18]王庆,张炜,赵荣椿.基于自适应门限和宽线检测的车牌定位方法[J],中国体视学与图像分析,2000(02):65-70.
[19]朱俊梅,陈少平.LPR系统车牌定位提取方法的研究[J],计算机工程与应用,2008,43(14):198-201.
[20]马永力,夏秋华.基于数学形态学的车牌精定位算法的研究[J];微计算机信息,2008,12(01):227-229.
[21]许智榜,石晓瑛.基于RBF神经网络的车牌自动识别系统设计与实现[J],南昌大学学报(工科版),2009,12(09):147-150.
[22]赵涛,杨晓莉,王绪本,张娜.一种用于车牌定位的改进BP神经网络方法[J],计算机仿真,2007,24(02):240-243.
[23]戴青云,余英林.一种基于小波与形态学的车牌图象分割方法[J],中国图形图象学报.2000,5A(5):411-415.
[24]王建平,姜涛.基于小波分析的汽车牌照分割[J],合肥工业大学学报(自然科学版)2002,25(6):1140-1142.
[25]侯培国,赵静,刘明.基于小波变换和行扫描的车牌定位方法[J].系统仿真学报,2006, 18(2):811-813.
[26]贾永红.数字图像处理[M].武汉:武汉大学出版社,2003:98-130.
[27]毕厚杰.新一代视频压缩编码标准--H.264/AVC,北京:人民邮电出版社,2006:35-59.
[28]魏诚.车牌图像二值化效果的改进算法研究[D] .浙江大学硕士学位论文, 2006.
[29]江萍,徐晓冰,方敏.基于Top-hat变换的车牌快速搜索与定位[J],合肥工业大学学报(自然科学版),2003,26(03):337-339.
[30]侯阿临,徐欣,史东承,陈勇.基于Top-Hat预处理和小波能量分析的车牌定位算法[J],吉林大学学报(信息科学版) , 2007,12(03):112-117.
[31] Mallat S, Zhong S. Characterization of signals from multi-scale edges[J], IEEE Transactions on PAMI,1992,12(07):710 - 732 .
[32] Yu.Zhong, Kalle.Karu, AnilK.Jain. Locating text in complex color images[J], Pattern Recognition, 1995,28(07):146-149.
[33] Xueming Qian, Guizhong Liu, Huan Wang, Rui Su. Text detection, localization, and tracking in compressed video[J],2007,22(09):752-768.
[34]冈萨雷斯等.数字图像处理[M],北京:电子工业出版社,2007:103-139.
[35]孙即祥.图像分析[M] .北京:科学出版社, 2005 : 186-211.
[36] Donoho D L , Johnstone I M. Adapting to Unknown Smoothness Via Wavelet s Shrinkage [J] . Journal of the American Statistical Association , 1995, 90 (432) : 1200-1224.
[37]黄山.车牌识别技术的研究和实现[D] .四川大学博士学位论文, 2005.
[38] Otsu N. A threshold selection method from gray-level histogram [J].IEEE Transaction on System Man and Cybernetic, 1979, 9(1) :62-66.
[39] Pun T. A New Method for Gray Level Picture Thresholding Using the Entropy of the Histogram [J]. Signal Processing, 1980, 2 (3):223-237.
[40] Kapur J N , Sahoo P K, Wong A KC. A New Mthod of Gray -level Picture Thresholding Using the Entropy of the Histogram [J]. Computer Vision Graphics and Image Processing, 1985, 29 (2):273-285.
[41] W H Tsai. Moment-preserving Thresholding:A New approach[C],CVGIP,1985:29:377-393.
[42] Kitter J. Minimum error thresholding[J] . Pattern Recognition, 1986, 19 (1):41-47.
[43] Chow C K, Kaneko T. Automatic detection of the left ventricle from cineangiogram [J ] . Computers and Biomedical Research, 1972, 5:388-410.
[44] Mardia K V , Hainswo rth T J. A Spatial Threshoding Method for Image Segmentation [J ]. IEEE Trans Pattern Analysis and Machine Intelligence, 1988, 10 (6): 919-927.
[45] Trier O D, Taxt T. Evaluation of Binarization Methods for Document Images[J].IEEE Transactions on Pattern Analysis and Machine Intelligence, 1995,17(3):312-315.
[2] Barroso,j, Dagless,E.L, Rafael,A, Bulas-Cruz,j. Number plate reading using computer vision Industrial Electronics[C],Proc IEEE International Symposium on Industrial Electronics(ISIE),Portugal, 1997:761-766 .
[3] Charl Coetzee. Based Number Plate Recognition Systems[C],Proc IEEE International Symposium on Industrial Electronics,1998:605～610.
[4]张引,潘云鹤.彩色图象定位新方法[J],中国图象图形学报,2001,6(4):374-377.
[5]郭捷,施鹏飞.基于颜色和纹理分析的车牌定位方法[J],中国图象图形学报,2002, 7(5):472-475.
[6]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,2004,12(02):204-208.
[7]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):224-225.
[8]马永力,夏秋华.基于数学形态学的车牌精定位算法的研究[J],微计算机信息,2008,24(01):227-229
[9]王锋,彭国华,赵强.一种快速、高效的车牌定位方法[J].计算机工程与应用,2007,43(11):120-123.
[10]张向东,马月涛,杨奇泽.一种基于灰度图像的车牌快速定位算法[J].电子科技,2007,212(05):173-177.
[11]赵涛,杨晓莉,王绪本,张娜.一种用于车牌定位的改进BP神经网络方法[J],算机仿真,2007,24(02):240-243.
[12]陈建坤,范春年.一种基于神经网络的车牌定位方法[J].辽宁工程技术大学学报, 2005,24(01):97-100.
[13]赵先军.基于神经网络的车牌识别技术研究[D].西安电子科技大学硕士论文,2005.
[14]杨晓敏.车牌识别关键技术研究[D],四川大学博士论文,2008.
[15]陈兆学,施鹏飞.基于灰度图像的车牌快速定位和分割方法[J],计算机工程.2006,32(09):173-177.
[16]韩永强,李世祥.基于线处理的子图像定位算法[J],计算机与现代化.1998(2):10-12.
[17]魏武,黄心汉,王敏.一种基于垂直字符边界特征的车牌定位方法[J],中国公路学报.2000,13(04):88-99.
[18]王庆,张炜,赵荣椿.基于自适应门限和宽线检测的车牌定位方法[J],中国体视学与图像分析,2000(02):65-70.
[19]朱俊梅,陈少平.LPR系统车牌定位提取方法的研究[J],计算机工程与应用,2008,43(14):198-201.
[20]马永力,夏秋华.基于数学形态学的车牌精定位算法的研究[J];微计算机信息,2008,12(01):227-229.
[21]许智榜,石晓瑛.基于RBF神经网络的车牌自动识别系统设计与实现[J],南昌大学学报(工科版),2009,12(09):147-150.
[22]赵涛,杨晓莉,王绪本,张娜.一种用于车牌定位的改进BP神经网络方法[J],计算机仿真,2007,24(02):240-243.
[23]戴青云,余英林.一种基于小波与形态学的车牌图象分割方法[J],中国图形图象学报.2000,5A(5):411-415.
[24]王建平,姜涛.基于小波分析的汽车牌照分割[J],合肥工业大学学报(自然科学版)2002,25(6):1140-1142.
[25]侯培国,赵静,刘明.基于小波变换和行扫描的车牌定位方法[J].系统仿真学报,2006, 18(2):811-813.
[26]贾永红.数字图像处理[M].武汉:武汉大学出版社,2003:98-130.
[27]毕厚杰.新一代视频压缩编码标准--H.264/AVC,北京:人民邮电出版社,2006:35-59.
[28]魏诚.车牌图像二值化效果的改进算法研究[D] .浙江大学硕士学位论文, 2006.
[29]江萍,徐晓冰,方敏.基于Top-hat变换的车牌快速搜索与定位[J],合肥工业大学学报(自然科学版),2003,26(03):337-339.
[30]侯阿临,徐欣,史东承,陈勇.基于Top-Hat预处理和小波能量分析的车牌定位算法[J],吉林大学学报(信息科学版) , 2007,12(03):112-117.
[31] Mallat S, Zhong S. Characterization of signals from multi-scale edges[J], IEEE Transactions on PAMI,1992,12(07):710 - 732 .
[32] Yu.Zhong, Kalle.Karu, AnilK.Jain. Locating text in complex color images[J], Pattern Recognition, 1995,28(07):146-149.
[33] Xueming Qian, Guizhong Liu, Huan Wang, Rui Su. Text detection, localization, and tracking in compressed video[J],2007,22(09):752-768.
[34]冈萨雷斯等.数字图像处理[M],北京:电子工业出版社,2007:103-139.
[35]孙即祥.图像分析[M] .北京:科学出版社, 2005 : 186-211.
[36] Donoho D L , Johnstone I M. Adapting to Unknown Smoothness Via Wavelet s Shrinkage [J] . Journal of the American Statistical Association , 1995, 90 (432) : 1200-1224.
[37]黄山.车牌识别技术的研究和实现[D] .四川大学博士学位论文, 2005.
[38] Otsu N. A threshold selection method from gray-level histogram [J].IEEE Transaction on System Man and Cybernetic, 1979, 9(1) :62-66.
[39] Pun T. A New Method for Gray Level Picture Thresholding Using the Entropy of the Histogram [J]. Signal Processing, 1980, 2 (3):223-237.
[40] Kapur J N , Sahoo P K, Wong A KC. A New Mthod of Gray -level Picture Thresholding Using the Entropy of the Histogram [J]. Computer Vision Graphics and Image Processing, 1985, 29 (2):273-285.
[41] W H Tsai. Moment-preserving Thresholding:A New approach[C],CVGIP,1985:29:377-393.
[42] Kitter J. Minimum error thresholding[J] . Pattern Recognition, 1986, 19 (1):41-47.
[43] Chow C K, Kaneko T. Automatic detection of the left ventricle from cineangiogram [J ] . Computers and Biomedical Research, 1972, 5:388-410.
[44] Mardia K V , Hainswo rth T J. A Spatial Threshoding Method for Image Segmentation [J ]. IEEE Trans Pattern Analysis and Machine Intelligence, 1988, 10 (6): 919-927.
[45] Trier O D, Taxt T. Evaluation of Binarization Methods for Document Images[J].IEEE Transactions on Pattern Analysis and Machine Intelligence, 1995,17(3):312-315.