International Journal of Computer Networking,

Wireless and Mobile Communications (IJCNWMC)

ISSN (P): 2250-1568; ISSN (E): 2278-9448
Vol. 8, Issue 4, Dec 2018, 1-8
© TJPRC Pvt. Ltd.

AUTOMATIC NUMBER PLATE RECOGNITION USING CANNY EDGE

DETECTOR, SLIDING CONCENTRIC WINDOWS AND

LINEAR DISCRIMINANT ANALYSIS

BRINTHAKUMARI. S1 & S. K. MUTHUSUNDAR2

1
Research Scholar, Computer Science and Engineering, Meenakshi Academy of Higher Education
and Research University, Tamil Nadu, India
2
Professor, Computer Science and Engineering, Sri Muthukumaran Institute of Technology,
Chennai, Tamil Nadu, India
ABSTRACT

This paper presents a novel methodology of detection and recognizing characters in number plates in vehicles
using Canny Edge Detector, Sliding Concentric Windows and Linear Discriminant Analysis. This framework of number
plate recognition plays a vital role in parking places, toll gates, traffic monitoring, security control etc. This framework is
implemented as a four step process. In the first step, the accepted image is preprocessed using Canny Edge Detector. This

Original Article
is a multi-stage algorithm that extracts wide variety of edges even in a noisy environment. The second step follows Sliding
Concentric Windows (SCWs) for plate detection. License plate is extracted based on the vertical and horizontal position
histogram and then morphological operations such as dilation and erosion are applied to get the right candidate plate
region. The third step is character segmentation using Connected Component Analysis (CCA). Final step is for character
recognition and it uses Linear Discriminant Analysis (LDA) using Local Binary Pattern (LBP) features. This work is
implemented on the UFPR-ALPR Dataset and the experimental results shows the recognition accuracy of 88.5%.

KEYWORDS: Canny Edge Detector, Sliding Concentric Windows, Linear Discriminant Analysis & Local Binary Pattern
Features

Received: Sep 04, 2018; Accepted: Sep 24, 2018; Published: Oct 27, 2018; Paper Id.: IJCNWMCDEC20181

INTRODUCTION

Automatic License Plate Recognition helps to identify vehicle in an efficient manner and with good
recognition accuracy, without the need for major human resources and has become more and more important the
recent years. There are several reasons for the importance of recognition of number plates in the vehicles. There are
a growing number of vehicles on the roads with fast moving rate that cannot be tracked efficiently. The rapid
development in image processing and computer vision technology has also made it possible to detect, identify and
recognize the numbers in the license plates at a fast rate.

License Plate Recognition systems have been implemented and automated strictly in many countries
throughout the world. The proposed work extracts the features like height and width of the plate, shape of the plate,
color and font of the characters embedded in the plate. Even if there exist standard pattern of number plates in India,
people follow their own pattern of styles. The variations are shown in terms of fonts, letter styles, font size, position
of characters, color of characters and number plates.

www.tjprc.org [email protected]
2 Brinthakumari. S & S. K. Muthusundar

The ultimate aim of this work is to design a deployable working system for Indian License Plate Recognition
which is a challenging task since there is no uniformity in the Indian License Plates, as each state has adopted its own style
and structure. In addition to the problems faced by the General License Plate Recognition Algorithms, there are some set of
specific problems needs to be solved by the Indian License Plate Recognition algorithms.

As there is inadequacy in the Indian Transportation Engineering and Infrastructure, the road rules are violated and
by the lack of monitoring capability, the number of accidents gets increased. The primary need is to identify the vehicles
using their license plates. But there prevails a non-uniformity in the Indian License plates and infrastructure.

The License plates may not be positioned in the correct position, according to the class of vehicles. This will lead
to the misclassification of plate image in the vehicle. Number plates of other countries are also misclassified. In highway
toll collection booths, the vehicles are queuing up for long hours and it becomes worst in peak hours to get the fee paid and
pass on the gates. By using license plate recognition systems, automatically the fee can be collected from the owners
account and no manual process in required in toll gates. The accidents are getting increased due to speed violation and
disobeying the traffic rules. These kinds of violations can easily be monitored by this system and such driving can be
penalized. Hence, if such systems are implemented, the best value from the road and rail systems can be prevailed and it
also helps in traffic flow more smoothly, reducing delays, fuel consumption, and air and noise pollution.

The proposed work is purely based on Indian License plates and hence an overview of the same is very much
essential. In India, by law, all license plates also known as number plate must be based on Hindu-Arabic numerals with
Roman alphabets. Being a multi-linguistic country, in many states the number plates will have local scripting too and
hence the Indian license plates are unique from other countries.

This paper is organized as follows. Related work is described in Section 2. Section 3 presents the data and
methods. Section 4 presents proposed work describing the process to detect the characters in the number plates of the
vehicles with good accuracy. Sections 5 presents the experimental results and Section 6 concludes the paper with some
recommendations for future work.

RELATED WORK

Several techniques exist for automatic detection of license plate.

In [1] Sarfraz. M Ahmed. M. J et al presents Automatic License Plate Recognition systems (LPR) which helps to
decrease the number of traffic violations and construct the streets safer. In that scheme they explained about an urbanized
automatic scheme that locates Saudi license plates in a captured image regardless of the time and the license plate scale.
The proposed system can stand slight tilting of the license plate. The localization process is complex due to the extreme
varying nature of the background content. Good results were obtained using the localization stage. A second division of the
system was developed for segmenting and distinguishing the characters in the positioned license plate. Locating the license
plate in a picture is the first velocity in any license plate recognition system.

In [2] Suri. Dr. P. K, VermaEr. A et al the presentation is about the numbers on vehicle plates which is used as the
input image and simple color conversion are used. Edge detection and connector measurement methods are also accounted
too. Throughout the whole work, they use the masking and smoothing operations. The median filter is worn as one of the
operators. The best consequences can be obtained by receiving the value of connector mechanism which is more than 17.
Various methods have been planned so far but there is always a difficulty to notice the numbers. The image is stored in the

Impact Factor (JCC): 7.9083 NAAS Rating: 3.71

Automatic Number Plate Recognition using Canny Edge 3
Detector, Sliding Concentric Windows and
Linear Discriminant Analysis
form of a matrix and the output is displayed in the shape of detected numbers. They use Sobel Edge Detection technique to
minimize the complexity (as in covariance matrix) and local variance scores have been used and the sole coefficients have
been efficient into a attribute vector form and multi-layer neural network. Since no explicit comparison or detachment
calculation is essential in this framework, it is very easily possible to maintain the computational weight of the detection
process low always.

S. Wang and H. Lee [3] combined edge statistics with morphological steps to eliminate unwanted edges in the
processed images. On the other hand, some approaches take advantage of color features. Lin Luo [4] creates a new and
effective approach of license plate location. The following three key sections are involved in this proposed algorithm.
Firstly, Sobel operator is used to extract the vertical edges of the vehicle image. Then, HSV color space and integral image
are employed to locate candidates in yellow license plates and non-yellow license plates. Finally connected component
analysis is used to locate the license plate accurately.

Similarly, Yao-quan [5] also uses a color-based method, but in which the color information from the colored
image is reasonably utilized to greatly decrease the edge points. With this method, it also eliminates the disturbances of the
fake plate’s region whose structure and texture are similar to the vehicle plate but do not match the plate’s fixed color
collocation.

The large contrast between the characters and the background is exploited in [6] to detect license plates with black
characters over white backgrounds. While some other algorithms assumed that the density of edges in the license plate
region is larger than other regions if the contrast of the character and the license plate is sufficiently large. For example, in
[7,8] they scanned the vehicle images with N-row distance to count the existent edges. Regions with high edge density will
likely have the license plate inside. Similarly, in [9], a block-based method was proposed, and blocks with high edge
magnitude and variance are considered as the license plate region.

Image transformation methods based on Hough transform, Gabor filters and wavelet transform have been applied
in license plate detection. Hough transform is a classic algorithm to detect straight lines. Since the shape of license plate
can be defined by lines, [10] used the Hough transform to detect the boundary of a license plate. This method is practical
only when the background of the image is simple. Another disadvantage of this method is that the computational
complexity of Hough transform is very high. Gabor filters are often used to analyze textures as they are sensitive to
textures with different directions and scales, data and methods.

DATASET – MEDIA LAB LPR DATABASE

This work is implemented on the benchmark dataset, English LP Database. This database contains 530 images of
640 ×480 resolution with various illuminations.

www.tjprc.org [email protected]
4 Brinthakumari. S & S. K. Muthusundar

Figure 1: Some Vehicle Images from Original Dataset

The entire flow of the proposed work is depicted in figure 2. The implementation comprises of four main steps
that includes (i) Preprocessing – Canny edge detector (ii) Plate detection using Sliding Concentric Windows (iii) Character
Segmentation using Connected Component Analysis (iv) Character Recognition using Linear Discriminant Analysis of
Local Binary Pattern features.

Figure 2: The Proposed Architecture of License Plate Recognition

METHODOLOGY
Canny Edge Detector

Edge in Vehicle Number plate provides very useful information for the recognition of number plate. Canny Edge
Detector extracts edge information for several edges. The standardized number plates followed in our nation are with the
following attributes: Seven characters in the plate, the size of each character is 90cm X 45cm, with 10cm space between
edge and character, 12cm spacing between the characters.

The Canny Edge Detector algorithm extracts wide variety of edges even in noisy environment and the pseudo
code is shown in figure 3.

Impact Factor (JCC): 7.9083 NAAS Rating: 3.71

Automatic Number Plate Recognition using Canny Edge 5
Detector, Sliding Concentric Windows and
Linear Discriminant Analysis

Figure 3: Algorithm of Canny Edge Detector

Sliding Concentric Windows Algorithm

In the second step of License plate detection, sliding concentric windows method is used. The task of license plate
detection is carried out in two steps. In the first step, a concentric window (A) is scanned from left to right and another
concentric window (B) is scanned from top to bottom. Then features like mean and standard deviation of A and B are
computed. Then evaluate the ratio between the two. If the ratio is greater than a threshold value, the central pixel of the
window indicates the Region of Interest (ROI). It is indicated as 1 for ROI otherwise 0.

In the second step, extract the intensity component. From the intensity feature explore the green, yellow and white
component values using Hue, Saturation and Intensity values.

Then generate the histogram for the license plate. And using the vertical and horizontal position of histogram
generate the exact plate portion of the license plate.

Finally, Dilation and Erosion are applied to extract the right plate region.

Figure 4: Algorithm of Dilation

Connected Component Analysis

Connected Component Analysis, CCA, is one of the most widely selected algorithms for the initial step of
character recognition in various segmentation methods. The connected component operator performs the scan from left to

www.tjprc.org [email protected]
6 Brinthakumari. S & S. K. Muthusundar

right till it reaches a point p, for which the ROI is 1. Then find the four neighbors of p. Check whether if all the four
neighbors are 0, then assign a new label to p. If only one neighbor is 0, then assign it to p. If more than one neighbors have
1, then assign 1 as label of p. Connected components labeling scans an image and groups its pixels into components based
on pixel connectivity.

Linear Discriminant Analysis

Each character candidate obtained from the previous segmentation phase comes with an aspect ratio ra and
orientation ∠LM

If ra is not in the range [1.6, 2.5] or ∠LM is not within [80◦, 100◦ ], showing that the character can be with a large
viewpoint rotation, an affine transform would be applied to warp it to within the desired ranges of ra and ∠LM.

The character candidate with appropriate aspect ratio and orientation is normalized to 30 × 15 pixels in scale and
recognized by a hierarchical classifier with two processing layers. The first layer has 25 classes that cover all 35
alphanumeric characters1 and one class that covers the non characters caused by plate edges, badly segmented characters
or segments from cluttered backgrounds, giving a total of 26 classes. Out of the 25 alphanumeric classes, seven classes
have an enhancement classifier in the second layer that targets a set of characters with high misclassification likelihoods.
The seven sets are [B, R, 6, 8], [C, G], [D, 0], [E, F], [I, 1], [U, V], and [Z, 2, 7]. The remaining 18 classes in the first layer
each identifies a specific character. The selection of the seven easy-to-misclassify sets is based on the classification results
from the experiments that only use the first-layer classifier, which was originally designed to identify the 35 classes of
alphanumeric characters.

Figure 5: Algorithm of LDA Classifier

EXPERIMENT EVALUATION AND RESULTS

Using the benchmark dataset, English LP Database 200 images are taken as training samples and 150 images for
testing and the rest for validation. The dataset is passed on the four modules sequentially one after the other. The work is
implemented using MATLAB and tested in Windows PC with Pentium Dual Core 2.4-GHz processor.

The accuracy of Detection, Segmentation, Recognition of characters along with time taken for detection in
Number Plate is specified in Table 1.

Impact Factor (JCC): 7.9083 NAAS Rating: 3.71

Automatic Number Plate Recognition using Canny Edge 7
Detector, Sliding Concentric Windows and
Linear Discriminant Analysis
Table 1: Accuracy of Character Detection in License Plate
Accuracy (%)
Letter Test Sample Time(s)
Detect Segment Recog Overall
0 210 90 89 85 88.2 0.06
1 225 80 90 80 87.6 0.12
2 300 78 89 78 78.6 0.10
3 291 89 88 86 84.6 0.09
4 223 79 88 75 80.2 0.12
5 212 89 85 85 81.5 0.08
6 235 88 80 86 81.6 0.09
7 247 78 79 70 79.5 0.11
8 301 84 86 92 87.4 0.13
9 201 95 92 85 88.1 0.16
TN 112 94 93 96 88.5 0.10

The performance result shows the proposed work of detection of number plate outperforms other existing
methods.

CONCLUSIONS

This work is implemented successfully with Canny Edge Detector for edge detection even in noisy environment,
Sliding Concentric Windows (SCWs) with dilation and erosion to extract the number plate efficiently, Connected
Component Analysis (CCA) for character segmentation and Linear Discriminant Analysis (LDA) using Local Binary
Pattern (LBP) features for character recognition. This work is implemented on the English LP Database Dataset and the
experimental results shows the recognition accuracy of 88.5%.

In near future, other interest region detectors deserve to be studied for their potential in improving license plate
recognition performance. The next version of the proposed work can include video samples of vehicle number plate for
recognition with good accuracy.

REFERENCES

1. M. Rasooli, S. Ghofrani and E. Fatemizadeh, “Farsi License Plate Detection based on Element Analysis and Characters
Recognition”, International Journal of Signal Processing, Image Processing and Pattern Recognition, vol. 4, no. 1, (2011)
March, pp. 65-80.

2. M. I. Khalil, “Car Plate Recognition Using the Template Matching Method”, International Journal of Computer Theory and
Engineering, vol. 2, no. 5, (2010) October, pp. 683-687.

3. S. Wang and H. Lee, “Detection and recognition of license plate characters with different appearances,” IEEE Intelligent
Transportation Systems, vol. 2, 2003.

4. L. Luo, H. Sun, W. Zhou and L. Luo, “An Efficient Method of License Plate Location” Information Science and Engineering
(ICISE), 2009 1st International Conference on. Page: 770. 26-28 Dec. 2009.

5. Y. Yang, J. Bai, R. Tian, N. Liu, “ A vehicle license plate recognition system based on fixed color collocation,” Machine
Learning and Cybernetics, 2005. Proceedings of 2005 International Conference on, Page: 5394, Issue Date: 18-21 Aug. 2005.

6. P. Comelli, P. Ferragina, M. N. Granieri, and F. Stabile, "Optical recognition of motor vehicle license plates," Vehicular
Technology, IEEE Transactions on, vol. 44, pp. 790-799, 1995..

www.tjprc.org [email protected]
8 Brinthakumari. S & S. K. Muthusundar

7. Patel, A. M., Patel, A., & Patel, H. R. (2013). Comparative Analysis For Machine Learning Techniques Appliance On
Anomaly Based Intrusion Detection System For Wlan.

8. J. Cano and J.-C. Pérez-Cortés, Vehicle License plate segmentation in natural images vol. 2652. Berlin, ALLEMAGNE:
Springer, 2003.

9. J. Barroso, E. L. Dagless, A. Rafael, and J. Bulas-Cruz, "Number plate reading using computer vision," in Industrial
Electronics, 1997. ISIE '97., Proceedings of the IEEE International Symposium on, 1997, pp. 761-766 vol.3.

10. L. Hsi-Jian, C. Si-Yuan, and W. Shen-Zheng, "Extraction and recognition of license plates of motorcycles and vehicles on
highways," in Pattern Recognition, 2004. ICPR 2004. Proceedings of the 17th International Conference on, 2004, pp. 356-359
Vol.4.

11. V. Kamat and S. Ganesan, "An efficient implementation of the Hough transform for detecting vehicle license plates using
DSP'S," in Proceedings of the Real-Time Technology and Applications Symposium, 1995, p. 58.

Impact Factor (JCC): 7.9083 NAAS Rating: 3.71