2023年11月28日发(作者：上海租车平台哪个好)
伊朗车牌识别使用连接组件和聚类技术
H.R.艾因Moghassemi 伊斯兰阿萨德大学（西德黑兰）伊朗德黑兰
摘要
车牌识别系统（LPR），在许多应用发挥了重要作用，如访问控制，流量控制，被盗车辆
的检测。一个车牌识别系统可分为检测和识别阶段。对于车牌检测，有一些相关的建议和方
法，就是对于水平板块和垂直板块的检测。车牌的准确的定位是认识到连接的成分分析和聚
类技术研究。由于对车辆定位的是摄像头，车牌矩形可以旋转所以在许多方面都会产生倾斜。
因此倾斜检测和校正车牌就很有必要。在这项研究中一个有效的歪斜检测和识别方法是泽尼
克旋转和尺度小波矩特征不变法用于车牌字符识别。和以上不同的是演算法是在强光照条件
下，视角，位置，大小和颜色在复杂的环境中运行时处理车牌。 “成功是在各种条件整体
性能达到车牌用于车牌识别系统时的93.54％。
关键词：
车牌识别;倾斜检测;斜线改正;泽尼克和小波矩;旋转和尺度不变。
一，导言
LPR（车牌识别）是一种用于识别车辆牌照的图像处理技术。 LPR系统中使用于各种安全
和交通应用，例如在图1。 LPR系统是用来在网关的访问控制。在图1：当车辆到达大门
时，自动车牌识别系统“读”车牌字符，与预定义的列表比较，如果有一个匹配则打开大门。
“LPR系统是在1976年首次由英国的分公司在警察科学开发。原型LPR系统是在1979
年工作。自1994年以来，伊朗第一个研究LPR系统的工作，开始在伊朗大学科技和技术
（IUST）和控制交通总公司开始实行[1]。
车牌识别使用图像处理软件分析图像捕捉车辆和定位提取车牌; 然后用光学字符识别
（OCR）系统对车辆图像进行车牌字符识别。他们还利用在各种警察,军队和使用电子收费
payper道路和交通的分类活动或个人。 LPR可以用来存储由相机拍摄的图像以及一些车牌
字符和数字，并且存储驱动程序。 LPR系统使用红外照明或图像加工技术，让相机拍摄照
片处理。
车牌识别系统的软件部分运行于中央，可以连接电脑和其他应用程序或数据库。该软件进
行车牌识别需要6种算法，如下：
1
，车牌的位置：查找和提取在图像上的车牌。
2
，车牌定位和缩放：调整的车牌歪斜度到所需的角度和大小尺寸，然后进行特
征提取和识别。
3
，正常化和二值化：调整车牌图像的亮度和对比度，将车牌图像转换为合适
阈值的二进制图像。
4
，字符分割：对车牌上各个字符分割开来。
5OCR
，光学字符识别（）：认识到车牌图像字符的特征。
6
，句法和几何分析：检查车牌字符和位置。
这些算法的复杂性决定了车牌识别系统整体的精度。本文的其余部分安排如下：在第2节，
提出有关工程在LPR的所有阶段与相关文献进行了系统的介绍和讨论。车牌预处理在第3
节。车牌检测算法是 LPR的基础和初级阶段，在第4节详细讨论。 第5节是关于车牌倾
斜检测校正。车牌字符识别在第6节，最后，实验结果提出在第7节。
二，有关工程
各种关于LPR系统的研究和工程应用，如停车，安全的访问和控制机密地区，交通执法等
可以在[1] [2] [3] [4] [5] [6]看到。如今研究集中到在没有统一的室外光照条件下各种车牌格
式，在图像采集，如背景，光照，车辆的速度，相机之间的距离不同。因此，大多数的方法
都有工作限制条件，如固定照明，有限车辆行驶速度，选择的路线，和固定的背景。上一节
所述的车牌识别系统由四个主要阶段构成：捕捉车辆的形象，定位和提取的车牌，字符分割
和正常化，光学字符识别。采集捕捉图像由相关硬件和摄像头完成。车牌定位和分割的LPR
系统的重要阶段。由于各种光照条件和复杂的背景，以寻找车牌的地方应予以考虑。还有噪
音，污染等可能会影响车牌的识别。关于定位识别车牌的许多研究基于边缘检测，遗传算法，
神经网络等。大多数建议的方法对车牌的亮度敏感，有很多的处理并没有强大到足以在各种
环境条件下发现车牌，独立地进行车牌识别 [7]。在这项研究中，计算车牌内的每一个字符，
并考虑为下一代阶段特征提取和识别。车牌字符识别，已经出现了波斯语/阿拉伯语大量的
光学字符识别技术。在本研究中一种新型的旋转不变泽尼克和小波矩的方法用于基地波斯语
/阿拉伯语光学字符识别[8]。
三，车牌预处理
本文是关于伊朗的一些样品车牌如图所示。车牌图像在不同的照明情况下被。
2
捕获，其中包括白昼，午夜，阴影和扭曲的条件。伊朗的车牌被归类为个人，政
府，公共服务和出租车，其中包括不同类型背景和前景。不同牌照的亮度可能会
有所不同，因为车牌的位置和
不同的照明环境。在预处理，应将车牌图像过滤将其转换
为二进制格式。将图像转换为二进制完成全局和局部阈值。由于全局阈值在不同光照条件下
不能总是产生满意的结果，所以合适的方法是局部阈值转换车牌为二进制格式的图像。在这
种情况下，车牌图像分为，个子图像，然后每个子图像局部阈值转换为二进制。在这
MN
项研究中算法用于计算局部阈值，。
OTSU[9][10]
四，车牌检测
车牌检测是系统的一个重要阶段，建议的方法应与前面的章节不同。建议
LPR
进行两个阶段进行。在第一阶段车牌图像上没有用处的地区进行出车牌夹角水平
和垂直预测。图显示了车牌图像水平和垂直方向的预测。垂直和图像的水平推
3
算，是一维信号，这代表了图像的整体幅度分别根据轴和轴。垂直和水平
YX
按照定义的预测方程：
其中，和是减值的。在第二使用阶段，连接成分分析和聚类上段的车牌字符。连接的
WH
组件分析（）是一个众所周知的图像处理技术，扫描图像和标签的像素成组件基于像
CCA
素和以某种方式相互连接（无论是四连接或八连接的）。面积，质心和边界框连接使用集群。
层次聚类程序是最常用的方法分组连接的部件。连接的结果成分分析和聚类样本图像车牌如
图所示。
3
五，倾斜检测与校正
由于对车辆定位的是摄像头，所以车牌的矩形可以旋转因此通常都会倾斜。由于偏斜大大降
低识别能力，所以要实施额外机制检测和纠正偏斜车牌。这种方法的基本问题是确定一个角
度，根据车牌歪斜。歪斜检测，第一次与连接成分分析计算车牌的所有字符的重心在图。
4
左边和右边的字符，然后与协调的倾斜角度计算如下：
在图左边和右边的字符，然后相关的例句协调的倾斜角度计算如下：
4
图显示斜度修正后的车牌。
4.B
六，车牌字符识别
为了识别一个车牌的字符，提取描述这个车牌是很有必要的。由于提取方法影响整个
OCR
过程的质量，这是非常重要的是提取功能，这将是对各种光线条件下不变的，使用的字体类
型和变形造成的图像歪斜的字符。第一步是亮度和对比度的正常化处理后的图像。包含字符
图像分割，调整合适尺寸（第二阶段）。之后，特征提取算法提取适当的归一化字符（第三
步）。在这项研究中，包含泽尼克和小波矩特征提取和前向神经网络的分类和识别。
[8]
七，实验结果
完整的测试图像数据库由个数字图像组成。多数图像代表伊朗的车牌在各种照明条件
1000
下获得的场面。车牌识别的整体成功性能在各类条件下达到。
93.53%
参考文献
[1]A. BroumandniaIUST1995
。，车牌自动使用数字图像处理“，硕士的认可。论文中，，
年。
“应用波斯语车牌模式识别承认“，杂志，。，法特希，
ICGST GVIP[2] A. BroumandniaM.
第卷，年月。
5Issue220051
[3] giannoukosAnagnostopoulosC.-N.Loumoskayafas
。，一，，，，五，，大肠杆菌，运
营商背景扫描，以支持高分割实时车牌识别率，模式识别，第卷第期，年，页
43112010
3866-3878
[4] AnagnostopoulosC.N.E.AnagnostopoulosIELoumosKayafas
，，，，五，，大肠杆菌的
一个车牌识别智能交通系统应用的算法“（）智能运输系统，（），艺术。没有。
2006IEEE73
1688109377-391
，页。
[5] AnagnostopoulosC.-N.E.AnagnostopoulosIEPsoroulasLoumos
，，，，编号，，五，
（）智能运，大肠杆菌的车牌从静止图像和视频序列的识别：一项调查“
2008IEEEKayafas
输系统，（），艺术。没有。，页。
93 4518951377-391
“迈向跨国汽车车牌识别系统“，（夏皮罗，五，研究，拥有，，
2006[6]GluhchevDimovD
年），机器视觉和应用，（），第。
173173 - 183
[7] JMGuoYFLiuIEEE
和，车牌定位和反馈的自学习和混合字符分割二值化技术，车辆交
易技术，，号，年，年至年。
Vol.573200814171424
“快泽尼克的小波时刻，波斯语字符识别阿里，，
[8]BroumandniaJamshid Shanbehzadeh
“图像与视觉计算（）。
252007717-726
。“调查显示，超过图像阈值技术和定量性能评估“杂。塞兹金（年）
[9]B. Sankur2004
志的电子影像（）：。
131146-165
。“阈值选取从灰度直方图“的方法。。。人。数码。：津信行（）
IEEESYS 9[10]1979
62-66
。
图：车牌识别系统在网关的访问控制
1
（一）个人车牌（）政府牌照
b
（三）公共服务牌照（四）出租车的车牌
图：四类型伊朗车牌
2
图：使用投影信号和连接成分分析的车牌检测
3
图：车牌倾斜检测和校正
4
Iranian License Plate Recognition Using Connected Component and
Clustering techniques
H.R. Ain Moghassemi
Islamic Azad University (West Tehran)
Tehran, Iran
Abstract:
License Plate Recognition system (LPR) plays a significant role in many application such
as access control, traffic control, and the detection stolen vehicles. A LPR system can be
divided into the detection and recognition stages. For license plate detection, a proposal
method with to phase is used. At the first phase regions of around plate is clip out by help of
vertical and horizontal projections. Next accurate location of plate is recognizing by
connected component analysis and clustering techniques. Due to the positioning of vehicle
towards the camera, the rectangular of license plate can be rotated and skewed in many
ways. So skew detection and correction is requiring after plate detection. In this study an
efficient method is proposed to skew detection and recognition. Zernike and wavelet
moments features with rotation and scale invariant property are used to recognition of
license plate characters. Proposed algorithms are robust to the different lighting condition,
view angle, the position, size and color of the license plates when running in complicated
environment. The overall performance of success for the license plate achieves 93.54%
when the system is used to the license plate recognition in various conditions.
Keywords:
License Plate Recognition; Skew detection; Skew correction; Zernike and wavelet moments;
rotation and scale invariant.
I. INTRODUCTION
LPR (License Plate Recognition) is an image processing technology used to recognize
vehicles by their license plates. LPR systems are used in various security and traffic
applications, for example in fig.1. LPR system is used to the access-control in the gateway.
In the figure 1: while the vehicle reaches the gate, the LPR system automatically \"reads\" the
license plate characters, compares to a predefined list and opens the gate if there is a
match. The LPR system was made-up first in 1976 at the Police Scientific Developed
Branch in the England. Prototype of LPR systems were working by 1979. In Iran first
research and work on LPR system started in Iran University Science and Technology (IUST)
and Control Traffic Corporation since 1994[1].
LPR uses image processing software to analyze the images capture from vehicles and
extracts the license plate location;at the end an optical character recognition (OCR) system
is used on images to recognize the license plates characters on Vehicles. They also are used
by various police and military forces and as a method of electronic toll collection on payper-
use roads and classification the activities of traffic or individuals. LPR can be used to store
the images captured by the cameras as well as the characters and numbers from license
plate, with some arrangement to store a photograph of the driver. LPR systems use
infrared lighting or image processing technique to allow the camera to take the picture for
next processing.
The software part of the LPR system runs on central computer and can be connected to
other applications or database. The software requires six algorithms for recognizing license
plate as follows:
1. License plate location: Finding and extracting the license plate on the image.
2. License plate orientation and scaling: compensates for the skew of the license plate
and resize the dimensions to the required size to feature extraction and recognition.
3. Normalization and binarization: adjust brightness and contrast of the license plate
image and convert license plate image to binary image with suitable threshold.
4. Character Segmentation: finds the individual bounding box of characters on the license
plate.
5. Optical Character Recognition (OCR): Recognizing characters of license plate image.
6. Syntactical and geometrical analysis: Check license plate characters and positions.
The complexity of each of these algorithms determines the precision of the overall LPR
system.
The rest of this paper is organized as follows. In Section 2, related works is presented, in
which all stages of LPR systems with related references are introduced and discussed. The
license plate preprocessing is presented in section 3. The basic and primary stage of LPR,
license plate detection algorithm, is discussed in detail in Sections 4. The section 5 is about
license plate skew detection and correction. The license plate character recognition is
presented in Section 6, and finally, experimental results are presented in Section 7.
II. RELATED WORKS
Researches and works about the LPR systems in various applications such as access
control to parking, security control of confidential areas, and traffic law enforcement can
be found in [1], [2], [3], [4], [5], [6]. Nowadays researches concentrate to variety of license
plate formats and the no uniform outdoor illumination conditions during image acquisition,
such as backgrounds, illumination, vehicle speeds, different distance between camera and
vehicle. Therefore, most approaches work only under restricted conditions such as fixed
illumination, limited vehicle speed, selected routes, and stationary backgrounds. As
described previous section LPR system consists of four main stages: capture an image of
the vehicle, locating and segmentation of the license plate, character segmentation and
normalization, and optical character recognition. Capture image related to hardware of
image grabber and camera. The license plate location and segmentation are the important
stage of LPR system. Because of various illumination conditions and complex backgrounds
to finding place of license plate should be considered. Also noise, dirty and skew plate can be
affected on performance of this stage. Many researches about locating license plate
recognition based on edge detection, genetic algorithm, neural network and etc can be
found in [6]. Most of proposal method to find location plate sensitive to brightness, have
more processing time and not robust enough to the various environment conditions. After
finding license plate, LPR system processing image of license plate and segment each
character individually[7].In this study, bounding boxes of all of characters are calculated and
considered for next stage to feature extraction and recognition. For recognition of license
plate characters, there have been a large number of Persian/Arabic optical character
recognition techniques. In this research a novel and rotation invariant method base on
Zernike and Wavelet moments is used for Persian/Arabic optical character recognition [8].
III. LICENSE PLATE PREPROCESSING
This paper is about Iranian license plate and some samples are shown in fig. 2. The images
of license plate are captured by different illumination which includes daylights, midnight, and
shadow and distortion conditions. Iranian license plates are categorized to personal,
governmental, public service and taxi types which include different background and
foreground. The brightness in a license plate may be vary because of the location of plate
and different of lighting environment. In preprocessing, license plate image should be
improved with suitable filtering and convert it to binary format. Convert image to binary can
be done with global and local thresholding. Since global thresholding cannot always generate
satisfied result in different illumination condition, an adaptive method with local thresholding
should be used to convert license plate image to binary format. In this case, license plate
image is divided into M by N sub images, and then each sub image is converted to binary by
local thresholding. In this study OTSU algorithm is used for calculate local threshold [9],[10].
IV. LICENSE PLATE DETECTION
The license plate detection is an important stage of LPR system, proposal approaches
should be worked with different image and various condition which is described in previous
sections. The proposal approaches is performed in two phases. In first phase uninteresting
regions of around license plate are clip out by horizontal and vertical projections. Figure 3
shows an image of license plate with horizontal and vertical projections. The vertical and
horizontal projections of an image is a one dimensional signal, which represent an overall
magnitude of the image according to axis y and x respectively. The vertical and horizontal
projections can be defined by following equations:
Where w and h are diminutions of the image. In second phase, connected component
analysis and clustering is used to segment characters on the license plate. Connected
components analysis (CCA) is a well-known technique in image processing that scans an
image and labels its pixels into components based on pixel and are in some way connected
with each other (either four-connected or eight connected). The area, centroids, and
bounding box of connected are used to clustering. Hierarchical clustering procedures are
the most commonly used method of grouping connected components. The result of
connected component analysis and clustering for a sample image of license plate is shown in
figure 3.
V. SKEW DETECTION AND CORRECTION
The rectangular of license plate can be rotated and skewed in many ways due to the
positioning of vehicle towards the camera. Since the skew considerably degrades the
recognition abilities, it is important to implement additional mechanisms, which are able to
detect and correct skewed license plates. The basic problem of this method is to determine
an angle, under which the license plate is skewed. For skew detection, first with connected
component analysis compute centroids of all characters in license plate as shownin figure 4.
Then with coordination’s of left and right REFERENCES characters compute skew angle
as follows:
To skew correction, apply transformation matrix A on license plate:
Figure 4.b shows a license plate after skew correction.
VI. THE LICENSE PLATE CHARACTER RECOGNITION
To recognize characters set of plate from a bitmap representation, there is a need to
extract feature descriptors of such bitmap. As an extraction method significantly affects
the quality of whole OCR process, it is very important to extract features, which will be
invariant towards the various light conditions, used font type and deformations of
characters caused by a skew of the image. The first step is a normalization of brightness
and contrast of processed image segments. The characters contained in the image
segments must be then resized to uniform dimensions (second step). After that, the
feature extraction algorithm extracts appropriate descriptors from the normalized
characters (third step). In this study, Zernike and wavelet moments are use to feature
extractions and feed forward neural network for classification and recognition[8]
VII. EXPERIMENTAL RESULTS
The complete testing image database consists of 1000 digital license plate images from
four sets. The majority of the images represent Iranian license plates from the natural
scenes obtained in various illumination conditions. The overall performance of success for
the license plate achieves 93.54% when the system is used to the license plate recognition in
various conditions.
REFERENCES
[1]. A. Broumandnia, ‘Automatic License Plate Recognition Using Digital Image Processing’,
MS. Thesis, IUST, 1995.
[2]. A. Broumandnia, M. Fathi,’ Application of pattern recognition for Farsi license plate
recognition’, ICGST-GVIP Journal, Volume 5, Issue2, Jan. 2005.
[3]. Giannoukos, I. , Anagnostopoulos, C.-N., Loumos, V., Kayafas, E. ,’Operator context
scanning to support high segmentation rates for real time license plate recognition’,
Pattern Recognition ,Volume 43, Issue 11, 2010, Pages 3866-3878
[4] Anagnostopoulos, C.N.E., Anagnostopoulos, I.E., Loumos, V., Kayafas, E. ‘A license
plate-recognition algorithm for intelligent transportation system applications ‘,(2006) IEEE
Transactions on Intelligent Transportation Systems, 7 (3), art. no. 1688109, pp. 377-391.
[5] Anagnostopoulos, C.-N.E., Anagnostopoulos, I.E., Psoroulas, I.D., Loumos, V., Kayafas,
E. ‘License plate recognition from still images and video sequences: A survey ‘, (2008) IEEE
Transactions on Intelligent Transportation Systems, 9 (3), art. no. 4518951, pp. 377-391.
[6] Shapiro, V., Gluhchev, G., Dimov, D.,’Towards a multinational car license plate
recognition system ‘, (2006) Machine Vision and Applications, 17 (3), pp. 173-183.
[7] and , License Plate Localization and Character Segmentation With
Feedback Self-Learning and Hybrid Binarization Techniques, IEEE transaction on vehicular
technology, Vol.57, No.3, 2008, 1417-1424.
[8] Ali Broumandnia, Jamshid Shanbehzadeh, ‘Fast Zernike wavelet moments for Farsi
character recognition’, Image and Vision Computing 25 (2007) 717–726.
[9] M. Sezgin and B. Sankur (2004). \"Survey over image thresholding techniques and
quantitative performance evaluation\". Journal of Electronic Imaging 13 (1): 146–165.
[10] Nobuyuki Otsu (1979). \"A threshold selection method from gray-level histograms\".
IEEE Trans. Sys., Man., Cyber. 9: 62–66.- 208 –
Figure 1: LPR system for access control in the gateway
(a) Personal license plate (b) Governmental license plate
(c) Public service license plate (d) Taxi license plate
Figure 2 : Four type Iranian license plate
Figure 3 : License plate detection using projection signals and connected components
analysis
Figure 4 : The License plate skew detection and correction
更多推荐
车牌识别
发布评论