a novel approach for car license plate detection based on

4
A Novel Approach for Car License Plate Detection Based on Vertical Edges Ashwathy Dev Department of Computer Science and InformationSystem, FISAT Angamaly, India [email protected] AbstractLicense plate can be used for identifying vehicle since it is unique for each vehicle.In this paper proposes a fast technique for identifying the vehicle licenseplate. Here first the input image is binarized by adaptive thresholding and then image is enhanced by unwanted-line elimination algorithm (ULEA). Then on applying VEDA vertical edges of the image is detected. Then number of possible candidate license plate region is extracted out of which original LP is detected. Keywords- binarized; adaptive thresholding; ULEA; VEDA I. INTRODUCTION The car-license-plate detection and recognition system have a number of applications, such as crime prevention, traffic data collection, and the payment of parking fees. For example in parking, number plates are used to calculate the duration of the parking. When a car reaches the gate, license plate is recognized automatically and stored in database. On leaving, the license plate is recognized again and compared with the stored numbers in the database. The time difference is used for calculating the parking fee. LPR is convenient and cost efficient as it is automated. Usually, a Car License Plate Detection Recognition System consists of three parts: license-plate (LP) detection (LPD), character segmentation and recognition. Among these, LPD is the most major part in the system because it affects the system’s accuracy. II. RELATED WORK There are number of LPD methods that have been used earlier, such as edge extraction [1], morphological operations [2], a neural network for colour [3], combination of gradient features [4], salience features [5], vector quantization [6], and grayscale [7] classification. In the past years, work was done on license plate detection in complex conditions. Kim et al. [4] proposed license plate detection algorithm using statistical features and license plate templates. In [8], [9], vertical edges of vehicle images are extracted by image enhancement and sobel operated. Here they remove most of the noisy edges. Finally using rectangular window in edge image the license plate region is searched. Ahmadyfard [10] have enhanced the method proposed in [9] by improving the quality of input image and then extract the vertical edges. Then morphological filtering is used to find plate regions. Zhang et al. [11] defined a vertical gradient map in order to extract statistical features. The researchers created two cascade classifiers based on Haar and statistical features to improve the detection rate and decrease the complexity of the system. In [12] feed forward neural network has been trained with the English character. Kim et al.[13] proposed an LPD algorithm using both statistical features and LP templates. After the statistical features were used to select the regions of interest (ROIs), LP templates were applied to match the ROI. III. PROPOSED METHOD FOR CAR LICENSE PLATE DETECTION SELECTING A TEMPLATE Figure 1.Flowchart of the proposed method Color input image Pre-processing Gray image conversion Adaptive Thresholding (AT) Median filter Vertical edges detection Unwanted-Lines Elimination Algorithm (ULEA) Vertical Edge Detection Algorithm (VEDA) Car license plate extraction Candidate Region Extraction (CRE) Plate Detection Plate Region Selection Display output image 2015 Fifth International Conference on Advances in Computing and Communications 978-1-4673-6994-7/15 $31.00 © 2015 IEEE DOI 10.1109/ICACC.2015.62 393 2015 Fifth International Conference on Advances in Computing and Communications 978-1-4673-6994-7/15 $31.00 © 2015 IEEE DOI 10.1109/ICACC.2015.62 391 2015 Fifth International Conference on Advances in Computing and Communications 978-1-4673-6994-7/15 $31.00 © 2015 IEEE DOI 10.1109/ICACC.2015.62 391

Upload: others

Post on 05-Feb-2022

3 views

Category:

Documents


0 download

TRANSCRIPT

A Novel Approach for Car License Plate Detection Based on Vertical Edges

Ashwathy Dev Department of Computer Science and InformationSystem, FISAT

Angamaly, India [email protected]

Abstract—License plate can be used for identifying vehicle since it is unique for each vehicle.In this paper proposes a fast technique for identifying the vehicle licenseplate. Here first the input image is binarized by adaptive thresholding and then image is enhanced by unwanted-line elimination algorithm (ULEA). Then on applying VEDA vertical edges of the image is detected. Then number of possible candidate license plate region is extracted out of which original LP is detected.

Keywords- binarized; adaptive thresholding; ULEA; VEDA

I. INTRODUCTION The car-license-plate detection and recognition system

have a number of applications, such as crime prevention, traffic data collection, and the payment of parking fees. For example in parking, number plates are used to calculate the duration of the parking. When a car reaches the gate, license plate is recognized automatically and stored in database. On leaving, the license plate is recognized again and compared with the stored numbers in the database. The time difference is used for calculating the parking fee. LPR is convenient and cost efficient as it is automated.

Usually, a Car License Plate Detection Recognition System consists of three parts: license-plate (LP) detection (LPD), character segmentation and recognition. Among these, LPD is the most major part in the system because it affects the system’s accuracy.

II. RELATED WORK There are number of LPD methods that have been used

earlier, such as edge extraction [1], morphological operations [2], a neural network for colour [3], combination of gradient features [4], salience features [5], vector quantization [6], and grayscale [7] classification.

In the past years, work was done on license plate detection in complex conditions. Kim et al. [4] proposed license plate detection algorithm using statistical features and license plate templates.

In [8], [9], vertical edges of vehicle images are extracted by image enhancement and sobel operated. Here they remove most of the noisy edges. Finally using rectangular window in edge image the license plate region is searched. Ahmadyfard [10] have enhanced the method proposed in [9] by improving the quality of input image and then extract the vertical edges. Then morphological filtering is used to find plate regions. Zhang et al. [11] defined a vertical gradient

map in order to extract statistical features. The researchers created two cascade classifiers based on Haar and statistical features to improve the detection rate and decrease the complexity of the system.

In [12] feed forward neural network has been trained with the English character. Kim et al.[13] proposed an LPD algorithm using both statistical features and LP templates. After the statistical features were used to select the regions of interest (ROIs), LP templates were applied to match the ROI.

III. PROPOSED METHOD FOR CAR LICENSE PLATE DETECTION SELECTING A TEMPLATE

Figure 1.Flowchart of the proposed method

Color input image

Pre-processing

Gray image conversion

Adaptive Thresholding (AT)

Median filter

Vertical edges detection

Unwanted-Lines Elimination

Algorithm (ULEA)

Vertical Edge Detection Algorithm

(VEDA)

Car license plate extraction

Candidate Region Extraction (CRE)

Plate Detection

Plate Region Selection

Display output image

2015 Fifth International Conference on Advances in Computing and Communications

978-1-4673-6994-7/15 $31.00 © 2015 IEEE

DOI 10.1109/ICACC.2015.62

393

2015 Fifth International Conference on Advances in Computing and Communications

978-1-4673-6994-7/15 $31.00 © 2015 IEEE

DOI 10.1109/ICACC.2015.62

391

2015 Fifth International Conference on Advances in Computing and Communications

978-1-4673-6994-7/15 $31.00 © 2015 IEEE

DOI 10.1109/ICACC.2015.62

391

A. Adaptive Thresholding The flow chart of the proposed method is shown in

Fig.1.First the color input image is transformed to grayscale, then median filter is applied which remove the noise. Then AT process is applied to get the binarized image.

The current pixel is matched with an average of neighboring S pixels. If the current pixel’s value is T percent lower than the average value, then the current pixel is set to black; otherwise, to white. Here one eighth of the image width is taken as S and T should be in the range 0.1 < T < 0.2 in our method.

First the integral image is computed. For that first find the sum of the values of pixel for every column jth through all row values i which is estimated using

( )th

th

jy

i

xj yxgi

==�=

0,)(sum (1)

here g(x, y) represents the input pixel values, and sum(i)|jth represents the cumulative gray values for the column jth through all the rows of the image I = 0, 1, . . . height.

Then, for every pixel the integral image can be computed as in (2): IntgrlImg(i,j)=

otherwisejif

isumjigIntgrlisum 0

),()1,(Im),( =

���

+− (2)

where IntgrlImg(i, j) represents integral image for pixel (i, j). In the next step do thresholding for each pixel. For that

first, compute the intensity summation for each window using two subtraction and one addition operation as follows:

sumwindow = ���

����

� ++

2,

2Im sjsigIntgrl

- ���

����

� −+

2,

2Im sjsigIntgrl

- ���

����

� +−

2,

2Im sjsigIntgrl

+ ���

����

� ++

2,

2Im sjsigIntgrl

(3)

where sumwindow represents the sum of intensities of the grayscale values for a particular local window. The boundaries is represented by

���

� ++

2,

2sjsi , �

��

� −+

2,

2sjsi , �

��

� +−

2,

2sjsi ,

and s represents the window size of the IntgrlImg, where s = image width/8. Therefore, in order to calculate the adaptive threshold value of the image, in which g(i, j) [0, 255] is the intensity of the pixel which is located at (i, j), threshold t(i, j) for each pixel is given by: ( ) ( ) windowsumTjit ×−= 1, (4)

here t(i, j) is the threshold for each pixel at location (i, j), and T is a constant, here, T = 0.15.

The threshold value of each pixel:

( ) ( ) ( )��� <×

=otherwise

jitSjigjio

,255,,,0

,2

(5)

here o(i, j) represents the adaptive threshold output and S2 represents the local window for the selected region.

B. ULEA In thresholding process generally it will produce many

unwanted thin lines which do not belong to license plate region. So ULEA is proposed in order to remove these unwanted lines in angles 00, 900, 450, and 1350 with width of one pixel. A 3x3 mask is used for all image pixels. Only black pixel values are tested. Consider g(x,y) represent the values for thresholded image. If the current pixel value is located at the center of the mask (x, y) is black, then test the 8-neighbor pixel values. If two corresponding pixel values are white in the same time, then convert the current pixel to white value. Fig. 2 shows all the cases in which the current pixel will be converted to background pixel. In the figure a, b, c and d represents possible white pixel location while passing the mask through the image.

Figure 2. The 4-cases for converting center pixel to background value, (a) for horizontal lines (0o), (b) for vertical lines (90o), (c) for inclined lines with (45o), (d) for inclined lines with (135o)

C. VEDA VEDA is applied to distinguish the plate detail region,

mainly the beginning and the end of each character. Therefore, the plate details can be detected easily. A 2 × 4 mask is used for this process, where the center pixel of the mask is located at points (0, 1) and (1, 1). The proposed mask consists of 3 smaller masks, which are left mask 2×1, center mask 2×2, and right mask 2×1. Fig.3 shows the design of the proposed mask, where x represents the rows or the height of the image and y represents the columns or the width of the image.

Figure 3. The design of the proposed mask: (a) moving mask, (b) left mask (0,0), (1,0), (c) center mask (0,1), (0,2), (1,1), (1,2), (d) right mask (0,3), (1,3)

394392392

The 2 × 4 mask starts moving in the imbottom and from left to right. If the four pix1), (0, 2), (1, 1), and (1, 2) are black, themask values if they are also black or not. If black, then the pixel at two locations at (0, be converted to white. Otherwise the pixel 1 will be taken. This process is repeated wthe image.

Figure 4. Flowchart of PRS and PD

D. CRE This process is mainly divided into 4 step1) Count the number of Drawn Lines peCount and store the number of lines that

per each row.

mage from top to xels located at (0, en test the other the all values are 1) and (1, 1) will value of column

with all pixels in

ps: er Each Row: have been drawn

2) Divide the Image into Multig In order to reduce the consume

into multi groups could be done. 3) Count and Store Group Index It is useful to use a threshold

groups and to keep the satisfied gdetails exist in.

4) Select Boundaries of CandidaThis step draws the horizontal

below each candidate region. Fig. 6

E. PRS This process aims to extract corre

plate region selection (PRS) and shown in Fig. 4.

IV. EXPERIMENTAL RESULT

Here, the image is applied on process will be evaluated. Finally,proposed CLPD method is evaluatethe CLPD method has been builVEDA CRE PRS PD.

Figure 5. Plate detection (a) input imamedian filter, (d) AT, (e) ULEA, (f) VEDA,

groups: ed time divide the image

xes and Boundaries: to eliminate unsatisfied

groups in which the LP

ate Regions: l boundaries above and shows output of CRE.

ect LP. The flowchart of plate detection (PD) is

TS AND DISCUSSION median filter then AT

, the performance of the ed. In the first evaluation, lt as follows: ULEA

age, (b) gray scale image, (c) , (g) CRE, (h) PD

395393393

Our proposed method has been tested on the laptop with the following specifications: Core i3 CPUs with 2 GHz and 2 GB of RAM. The program is running under Windows 8.1 written in MATLAB R2012a. Fig 5 shows different detection steps and fig. 6 shows final output.

Figure 6. License plate

Table 1 shows detection rate of the proposed and other several methods.

TABLE I. DETECTION RATE OF PROPOSED AND SEVERAL METHODS

[Reference] Detection Rate(%)

[7] 91.7 [12] 90 [13] 92.8

VEDA 96

V. CONCLUSION Automatic license plate recognition is quite challenging

due to the different license plate formats and the varying environmental conditions. There are numerous LPR techniques have been proposed in recent years. Some LPR systems can work under uncontrolled environment but some in restricted environment. We have proposed a fast algorithm for detection of vertical edge. We have proposed a LPD method in which we employed 50 images under different conditions. In the experiment, the rate of correctly detected LPs is 96%.

ACKNOWLEDGMENT This work is supported and guided by my research guide.

I am very thankful to my research guide Ms. Shimy Joseph, Assistant Professor, Computer Science Department, Federal Institute of Science and Technology (FISAT), India for her guidance and support.

REFERENCES [1] S.-H. Le, Y.-S. Seok, and E.-J. Lee, “Multi-national integrated car-

license plate recognition system using geometrical feature and hybrid pattern vector,” in Proc. Int. Tech. Conf. Circuits Syst. Comput. Commun., Phuket, Thailand, 2002, pp. 1256–1259.

[2] J.-W. Hsieh, S.-H. Yu, and Y. S. Chen, ‘‘Morphology-based license plate detection from complex scenes,’’ in Proc. 16th Int. Conf. Pattern Recognit.,Quebec City, QC, Canada, 2002, pp. 176---179.

[3] E. R. Lee, K. K. Pyeoung, and J. K. Hang, “Automatic recognition of a car license plate using color image processing,” in Proc. IEEE Int. Conf.Image Process., 1994, pp. 301–305.

[4] S. Kim, D. Kim, Y. Ryu, and G. Kim, “A robust license-plate extraction method under complex image conditions,” in Proc. 16th Int. Conf. PatternRecognit., Quebec City, QC, Canada, 2002, pp. 216–219.

[5] Z.-X. Chen, Y.-L. Cheng, F.-L. Chang, and G.-Y. Wang, “Automatic license-plate location and recognition based on feature salience,” IEEE Trans. Veh. Technol., vol. 58, no. 7, pp. 3781–3785, Sep. 2009.

[6] S. Rovetta and R. Zunino, ‘‘License-plate localization by using vector quantization,’’ in Proc. Int. Conf. Acous., Speech, Signal Process., 1999, pp. 1113---1116

[7] H. Caner, H. S. Gecim, and A. Z. Alkar, “Efficient embedded neural network based license plate recognition system,” IEEE Trans. Veh. Technol., vol. 57, no. 5, pp. 2675–2683, Sep. 2008.

[8] H.-H. P. Wu, H.-H. Chen, R.-J. Wu, and D.-F. Shen, “License plate extraction in low resolution video,” in Proc. IEEE 18th Int. Conf. Pattern Recognit., Hong Kong, 2006, pp. 824–827.

[9] D. Zheng, Y. Zhao, and J. Wang, “An efficient method of license plate location,” Pattern Recognit. Lett., vol. 26, no. 15, pp. 2431–2438, Nov. 2005.

[10] V. Abolghasemi and A. Ahmadyfard, “Improved image enhancement method for license plate detection,” in Proc. 15th Int. Conf. DSP, 2007, pp. 435–438.

[11] H. Zhang, W. Jia, X. He, and Q. Wu, “A fast algorithm for license plate detection in various conditions,” in Proc. IEEE Int. Conf. Syst., Man,Cybern., Taipei, Taiwan, 2006, pp. 2420–2425.

[12] R. Parisi, E. D. Di Claudio, G. Lucarelli, and G. Orlandi, Car plate recognition byneural networks and image processing, in Proc. IEEE Int. Symp. Circuits Syst.,1998, pp. 195198.

[13] S. K. Kim, D. W. Kim, and H. J. Kim, A recognition of vehicle license plateusing a genetic algorithm based segmentation, in Proc. Int. Conf. Image Process.,Lausanne, Switzerland, 1996, pp. 661664.

396394394