Vous êtes sur la page 1sur 6

Automatic license plate recognition

Mohamed El-Adawi Hesham Abd el Moneim Keshk


Helwan Faculty of Engineering Helwan Faculty of Engineering
mhhha@naseej.com h_keshk@mail.com

Mona Mahmoud Haragi


Helwan Faculty of Engineering
mona_haragi@hotmail.com

Abstract: Automatic license plate recognition a fuzzy artmap neural to cluster these basic
(LPR) is one of the techniques that can be used features into different classes, The clustering
for the identification of vehicles. The purpose of produced blobs that satisfy the original features,
this paper is to investigate a suitable way to which are considered as the characters of the
recognize the registration plate from an image of license plate. Xulio Fernandez and Fernando
vehicle, The proposed algorithm recognizes the martin [4] used the character location method to
Arabic (Hindi)digits and words within the plate license plate recognition, They designed an
area. The obtained recognition ratio is about 80% algorithm to locate all zones in the image that
for a 65 random images included in the database. may contain characters. Their model is based on
The algorithm failed only in images that include derivatives, where they computed the horizontal
abnormalities such as fixing screws, or a major Sobel gradient of the whole image and search for
slant in the plate. high positive values followed by a high negative
keywords: thresholding, edge detection, dilation, ones. They look for rectangular zones in which
watershed transform, template matching, OCR. they have a repetition of the pattern. The distance
1.Introduction between the positive values and the negative
There has been a number of software products ones must be approximately constant, Their
that can be used for LPR mainly from USA and algorithm tries to find the possible rectangular
Europe.V.Kamet and S.Ganesan [1] used the that enclosed the plate. Another algorithm
Hough transform (HT) for line detection to proposed by J.R.Cowell.[5] where he applied a
locate the boundary line of the vehicle license global threshold to the whole image and finds the
plate. They first applied a modified Sobel plate using template matching and growing
operator for edge detection and then an image regions techniques. This method required a fixed
thresholding is done depending on the average size for the plate area and also the same degree
brightness of the image. HT is applied to the of illumination, which can not be satisfied easily.
entire image, and peaks are searched in the Su-Hyun Lee,Young-Soo Seok and Eung-Joo
Hough space in which the lines are identified in Lee[6] proposed license plate recognition
real space. Color and texture of the license plate system for multinational vehicle license plate
have been used to identify the plate lines. using geometric features along with hybrid and
J.A.G.Nijhuis [2] incorporated fuzzy features of seven segment pattern vectors. In the proposed
the license plate, using yellow rectangular area system, they suggested to find horizontal and
and black characters. He used a discrete time vertical relation after going through preparation
cellular neural network (DTCNN) as the feature process with inputted real-time license plate
extractor and multi-layered perception (MLP) as image of Korea and Japan, and then to classify
the recognizer. Yap Keem Siah, Tay Yong Haur, license plate with using characteristic and
Marzuki Khalid and Tahir Ahmed [3] used an geometric information of license plates. It
iterative threshold to fully exploit the large classifies the extracted license plate images into
contrast between the black color and background letters and numbers, such as local name, local
and with color characters of the license plate, number, classification character and license
They used the blob analysis method to identify consecutive numbers, and recognize license
the connected regions of pixels in the image and plate of Korea and Japan by applying hybrid and
then the basic features(such as height, width, top seven segments pattern vectors to classified letter
and left) of the region is calculated. These and number region. David Chanson and Timothy
regions are commonly know as blobs, They used Roberts [7] used Image processing techniques
such as edge detection, thresholding and
resampling in their algorithm to locate and 2.The plate has dark character on a bright white
isolate the license plate and the characters. back ground.
Template matching was the character recognition 3.The width-height relationship of the license
method that was used to successfully recognize plate is approximately known in advance.
the license plate. Jaime José Laracuente-Díaz 4.The orientation of the license plate is
and Manuel Toledo-Quiñones [8] determined to approximately aligned with the axes(vertical and
partion the image that contains the license plate, horizontal).
the different objects present in the image are 5.Lines of image must not be optically distorted.
found and labeled. Those that have geometrical We have been working with color images
characteristics similar to those of license plate captured with a digital camera and transferred to
characters are selected. If the relative position of computer memory. Our images consist of a
the selected objects is consistent with a license rectangular array of 640*480 pixels. Fig.1 shows
plate, it is assumed that the plate has been found. a sample of such images.

2.LPR advantages and applications


One of the advantages of LPR systems is that all
cars already have or at least are supposed to have
the required identification which is the
registration plate. This means that there is no
need to add transmitters or special signs to the
vehicles, which in turn results in savings towards
the implementation of the system. Another
advantage is that the image of the car and may
be even the image of the driver can be stored and
retrieved if there is a need for evidence in
disputes or crime situations. LPR systems have
several applications ranging from allowing car Fig.1 Car image with recognizable LP
park entry to detection of stolen cars. Long term
parking at the airport is also a good candidate for 3.1 Extraction of the license plate
the use of an LPR system as the actual time of The purpose of this stage is to locate the plate
entry can be retrieved in the event of a lost ticket that encloses the license number. The whole idea
that was issued days ago or in case there is a depends on the boundary of the license plate
ticket fraud situation. In a border control only because there are a huge data in the image.
situation, the LPR system can be used to double The algorithm tries to decrease the amount of
check that the car is allowed to cross and shorten data in the image and avoid the unwanted areas
the time it takes to go through the border, hence to extract only the area which have the same
shortening the typical long queues. The LPR properties of the license plate. Our proposed
system can also be used as a clever way of algorithm goes as follow:
directing traffic at the entrance of an
organization’s property. For instance, vehicles 1- Thresholding of the input image.
entering could have their plates read and After turn the input color image to a grayscale,
they could be directed to the suitable lanes we compute a global threshold level that can be
depending on their privileges as specified in a used to convert a gray image to a binary image.
database. An elaborate LPR system can also be In a grayscale image, the thresholding
used to monitor traffic on the roads in order to transformation sets each gray level that is less
detect stolen cars or cars with unpaid fines. than or equal to a chosen value T ( the threshold
value ) to 0 and each gray level greater than T
3.Proposed VLP recognition algorithm to 1.The result is a black and white image with 0
In such algorithms, one of the most difficult representing black and 1 representing white. In
tasks is to locate the license plate, which could our algorithm the threshold value is not constant,
be any where in the image. We make the Iit depends on the average brightness of the
following assumptions about plates proposed in image. We used the otsu’s method [9] to
this algorithm : determine that value.Figs.2 and 3 show the
1.The plates have a rectangular shape with two applications of different threshold values to
rows of characters. convert the image into binary image.
4- Reject small lines that are impossible to be
a part of the plate borders in both directions.
For each point checking if there is a new point in
the same line within certain distance. If it is, the
line segment is extended to include the new point;
otherwise a new line segment is started. In this
Fig.2 The computed threshold is 127 project the maximum distance, was chosen as the
value which gave the best results for the training
images. Also, as some of the line segments are
potentially very short, specifying a minimum
line segment length is advantageous and helps
reduce computation in later steps. The long lines
segments from the image have been extracted.
The short lines segment in the letters and in the
top part have been eliminated. Fig.6 shows the
Fig.3 The computed threshold is 92 output of the line extract process.
2- Apply a thin edge detector
If an image consists of objects of interest on a
contrasting background, an edge is a transition
from background to object or vice versa. The
total change in intensity from background to
object is called the strength of the edge. There
are many methods of performing edge detection.
We detected the edges here by using the spatial
Fig.6 Image after limination of short lines and
filtering. We used convolution kernel for
extraction of long lines
horizontal direction [-1 1]and for vertical
direction is [-1;1].Those values will help us to
5- Now we have a group of lines in horizontal
detect any thin edge in the image as shown in
and vertical direction with different lengths. By
fig.4
adding the two images, we obtain many regions
with different areas. We used the watershed
transform to segment those regions each with it’s
properties. Fig.7shows the result of the adding .

Fig.4 Horizontal and vertical edge detector

3- Dilation the image.


The obtained image after edge detector is one
pixel thickness for each edge. If there is any
ripples on any line, then the line will appear as a
broken one. To avoid this problem we will make Fig.7 Adding image
a dilation for all lines in both horizontal and
vertical directions as shown in fig.5 The watershed transform is the method of choice
for image segmentation in the field of
mathematical morphology. The watershed
transform can be classified as a region-based
segmentation approach. The intuitive idea
underlying this method comes from geography,
It is that of a landscape or to pographic relief
which is flooded by water. Watersheds being the
divide lines of the domains of attraction of rain
falling over regions. An a alternative approach is
Fig.5 Dilation in both directions to imagine the landscape being immersed in a
lake, with holes pierced in local minima. Basins technique, which depends strongly on
(also called ‘catchments basins') will fill up with comparing each portion of the investigated
water starting at these local minima, and, at image to a template image(shown in fig.9).The
points where water coming from different basins true license plate in the image is found as the
would meet, dams are built. When the water region bearing the most resemblance to the
level has reached the highest peak in the template.
landscape, the process is stopped. As a result, the
landscape is partitioned into regions or basins
separated by dams, called watershed lines or
simply watersheds. When simulating this process
for image segmentation, two approaches may be
used: either one first finds basins, then
watersheds by taking a set complement; or one Fig.9 the proposed template image
computes a complete partition of the image into
basins, and subsequently finds the watersheds by A common way to perform this task is to use a
boundary detection. To be more explicit, we will cross correlation scheme[11].Fig.10(a) shows an
use the expression `watershed transform’ to example of a true plat be selected, and the
denote a labeling of the image, such that all corresponding a mesh of the cross correlation
points of a given catchments basin have the same factor where the maximum value is 18400.
unique label, and a special label, distinct from all Fig.10(b) shows an example of a true plat be
the labels of the catchments basins, is assigned to selected, and the corresponding a mesh of the
all points of the watersheds. Fig.8 shows the cross correlation factor where the maximum
output of the watershed transform [10]. value is 9000.
18400

Fig.10(a) Cross correlation with true plate


9000
1 2 3

Fig.8 watershed transform output

6- As mentioned, the output from watershed Fig.10(b) Cross correlation with false region
segmentation consists of a list of regions, of
which one hopefully contains the license plate. 3.2 character segmentation:
Two steps are performed in order to select the 1.The next step after the identification of the
best candidate. rectangular plate is to isolate its characters or
Area: The candidate region with small area or digit contents. All plate rectangle are assumed to
which have very large area will be voided, contain two rows of characters. For the image,
Height- width ratio: Although some times a we need to isolate the two lines as shown in
region with the license plate will contain some fig.11.
edges around the actual plate, the ratio provides a
mean of sorting out a lot of regions that could
not possibly contain a license plate.
7- In some cars we found that the back light
indicators have almost the same characteristics
of the plate (area and ratio between width and
height) as shown in fig.8(regions number 1 and
3). To distinguish between such false areas and
the true license plate (as shown in fig.8 region
number 2 ), we used a template matching Fig.11 the extract plate and the required isolation
Fig.14 The two types of plate registration
To do this we calculate the horizontal
summation of all black pixels in each row. The 4.Experimental results:
peaks of the summation will identify the two The system was tested for 65 images under
rows of characters as shown in fig.12. The two several conditions. The recognition ratio was
arrows point to the location of the number and very good (almost 80%).The 20%wrong
registration parts. identification came from the fixation screws of
the plate that was recognized as zero. If the
distance between camera and car was larger than
certain limit, the system fails to correctly identify
the plate. Fig 15(a)and fig.15(b)shows the
application of the proposed algorithm and the
response of the program was to identify the plate
number and its registration.

Fig.12 The horizontal projection


2.After isolating the two rows of characters
within the plate area, we take the digits row
alone and try to isolate each digit.
3.On the digits row we vertically count the
number of black pixels in each column.
This is actually a vertical projection. The peaks
of the projection will represent the existence of a
digit, and the valley will represent the isolation
or boundary between these digits. Fig.13 shows
this projection. Fig.15(a)

Fig.13 The vertical projection


3.3 Character identification:
The final task of the LPR is the recognition of
the plate contents after the isolation process in
the pervious step. We used a 3 layer neural
network trained with the back propagation
algorithm to identify the following objects in
each plate :- Fig.15(b)
1.The digits from 0 to 9.
2.The word " ‫" ﻣﻼ آـــﻲ‬. 5.Interpretation of results:
3.The two word "‫" اﻟﻘـﺎهﺮة‬and "‫" اﻟﺠﻴﺰة‬that exits in We test the algorithm on 65 various images, the
all cars that are registered in either Cairo or results are as follows:
El_giza. • 89% success rate for license plate
We used ten plate from the " ‫" ﻣﻼ آـــﻲ اﻟﻘـﺎهﺮة‬ recognition, the algorithm failed in
registration and another ten from the recognition the plates which has a major
" ‫ " ﻣﻼ آـــﻲ ﺝﻴـﺰة‬registration as the training set slant.
for the neural network. The remaining plate • 93% success rate for character
images were used for testing. The performance recognition of the extracted plates, the
of the neural network was very good. failed occurs for the plate which has
fixing screws in the character position.
• 96% success rate for character
recognition of all the characters on the
extracted plates, the illumination and
the blurring of the characters was the [7] License plate recognition system ,David
main reason of the failed cases. Chanson and Timothy Roberts,Department of
Electrical and Electronic Engineering ,Manukau
60Comparsion with other similar projects: Institute of Technology,Auckland.
Although all the mentioned methods deal with [8] Artificial Vision for Vehicle Monitoring at
the English letters and Arabic numbers which are Tren Urbano Stations Jaime José Laracuente-
mainly different with the characteristics of the Díaz and Manuel Toledo-Quiñones Electrical
Arabic letters and Hindi numbers, we can say and Computer Engineering Department
that our work has a good ratio (80%) against to University of Puerto Rico,Mayagüez Campus
(70% ) success rate mentioned in [7], (83.5 %) in Mayagüez,Puerto Rico 00681-5000
[6] and (74% ) in [8]. [9] ]N.Otsu,“A Threshold Selection Method
from Gray-Level Histograms,”IEEE
7.Conclusion Transactions on Systems,Man,and
Our system was able to identify license plate Cybernetics,vol.9,no.1,pp.62-66,1979
with a good accuracy. The source of errors was [10] Luc Vincent and Pierre Soille,“Watersheds
due to mainly the fixation screws of the plate, or in Digital Spaces:An Efficient Algorithm Based
due to a major slant in the plate axis. The on Immersion Simulations,”IEEE Transactions
algorithm will be extended to include different of Pattern Analysis and Machine
shape of plates with different registration. Intelligence,vol.13,no.6,June 1991,pp.583-598.
8.References: [11] J.P.Lewis,”Fast Normalized Cross
[1]Kamat ,V.and Ganesan ,S, ”An efficient Correlation”,Industrial Light&Magic.
implementation of the Hough transform for http://www.idiom.com/~zilla/Papers
detecting Vehicle license plates using DSP’S” /nvisioninterface/nip.html
Cellular Neural Networks and their application [12] Serra,J.,Image Analysis and Mathematical
proceedings. The IEEE 31st annual Morphology,Vol.2:Theoretical
1997International Carnahan Conference,1997,pp Advances.Newyork:Academic,1988
209-218. [13]Rafael C.Gonzalez,Richard
[2]Nijhuis J.A.G,Brugge ,M.H.,Helmholt E.Woods.”Digital Image
K.A.,”lincese plate recognition using DCTNNs,” Processing”1993,Addison-Wesley
Security Technology,1997,Proceedings 1998
fifth IEEE international workshop on
publish,1998,pp 212-217.
[3] Yap Keem siah ,Tay Yong Haur,Marzuki
Khalid and Tahir Ahmed , Vehicle licence plate
recognition by fuzzy artmap neural network ,
Centre for Artificial intelligence and robotics
(CAIRO) ,Faculty of Electrical
Enginerring ,University Teknologi
Malaysia,Jalan Lumpur ,54100 Kuala Lumpur.
[4] Xulio Fernández Hermida,Fernando Martín
Rodríguez,José Luis Fernández Lijó,Fidel Pita
Sande,Miguel Pérez Iglesias.Departamento de
Tecnologías de las
Comunicaciones.E.T.S.I.T.Universidad de Vigo.
Ciudad Universitaria s/n.36200 Vigo
(Pontevedra).Spain.”An O.C.R.for V.L.P.”
[5]J.R.Cowell.“Syntactic Pattern Recognizer for
vehicle identification Numbers”.Image
&Vision Computing (1995).
[6] Multi-National Integrated Car-License Plate
Recognition System Using Geometrical Feature
and Hybrid Pattern Vector Su-Hyun Lee,Young-
Soo Seok and Eung-Joo Lee Department of
information/Communication Eng.,TongMyong
Univ.of Information Technology

Vous aimerez peut-être aussi