Beruflich Dokumente
Kultur Dokumente
ABSTRACT
Car plate recognition is one of the most useful practical applications of optical character recognition. It can be applied
in many different scenarios, such as access control, parking fares calculation, automatic tolls and parking payment,
traffic control, etc. As other OCR systems, the neural network is the most used technique in character recognition
stage. Thus, most of Car plate recognition systems are suffering from neural network weakness which is the tradeoff
between accuracy rate, and the computation time. In this paper, a new technique of optical character recognition that
relay on Fast wavelets transform (FWT) is proposed. The idea of the proposed technique is to use the FWT to
produce a coefficient vector of the character images which will be directly used to recognize the characters. Due to
the vehicle registration plates of Malaysia, our approach can be used to detect Malaysian car plates.
1. I NTRODUCTION
Wavelet transform is a computation that localizes a function in both space and scale. It’s
known as discrete wavelets transform if the wavelets are discretely sampled. Mostly, DWT is
used for signal coding, to represent a discrete signal in a more redundant form. For analyzing, the
signal (S) is passed through filters (low and high pass filters in each level) (F) with different
frequencies at different scales. The output of this process (known as Decomposition) will be two
sets of coefficients: the approximations (A) which is high-scale, low-frequency components of
the signal, and the details (D) which is the low-scale, high-frequency components. Figure (1)
shows the output of passing the signal S through one level of the transform [14].
Signal
A D
Decomposition Reconstruction
F F
S S
F F
Cheng and LI Shao-Fa used simplified a pulse coupled neural network model with the
parameters determined by images’ spatial and grey characteristics automatically. Their work can
be considered as enhancement of Xin Yuan’s algorithm, which introduced the image
enhancement process based on a pulse coupled neural network model firstly. Using Xin Yuan’s
algorithm, a recognition rate of 98% can be achieved [7]. However, the neural network was used
by many researchers as recognition method, such as: [15], [16], and [17].
In this paper, we used the FWT to produce a coefficient vector of the character images. This
coefficient vector will be directly used to recognize the characters. Regarding to the vehicle
registration plates of Malaysia, our approach is suitable to be used to recognize Malaysian car
plates.
2. DISCUSSION
There are many techniques applied by different researchers in the recognition phase. The
neural network is the most used technique. Some of the researchers used multi-method to deal
with different types of characters used in car plates such as the combination of Chinese and
Roman characters or to deal with car plates with Arabic characters. While the multi-method
could improve the accuracy rate, it could effect the computation time which is very important in
any real-time car plate recognition. To circumvent the weakness of the artificial neural network
in order to get a high accuracy, we need to use many features that can provide enough
information, in this case we need to use complex artificial neural network that will need longer
response time. If we use a simple artificial neural network, the time will be reduced but the
accuracy will be reduced as well. Most of the reviewed papers did not report about the consumed
time which supposed to be long when neural network is used. The consumed time will be even
longer when combined methods are used [5], [6], and [13]. Our approach is to achieve the
recognition stage without using artificial neural network.
3. EXPERIMENTS
In this paper, we focus only on the recognition stage. The whole system, which includes the
design of the interface between the PC and the Camera and the preprocessing steps such as plate
localization, plate orientation and sizing, plate normalization, and character segmentation, could
be part of a future work.
In our experimental work, we aim to:
1- Proof that the decomposition coefficient vector (DCV) is unique for each character
2- This DCV can be used to recognize the image.
In [8], Gonzalea, Wood and Eddins developed a Matlab functions (wavefast and wavefilter) that
could generate wavelet decomposition and reconstruction filters. We are concerned only by the
decomposition part since we need only this part to get the DCV. Both functions were used in our
experimental work.
In our experimental work, we developed a Matlab function that achieves the segmentation
stage then calls a modified wavefast function to be applied on 40 pictures Figure (3) shows one
of the car pictures before preprocessing.
After preprocessing (plate detection) the car plate image is converted to binary image (this
step is important to reduce the noise) then the car plate image will be ready for segmentation
step.
Figure (4) shows the car plate (binary image) and the segmented number image
After segmentation, we obtained 10 images of each number (0-9). After applying wavefast
function, we got the DCV of the ten numbers; to avoid overlapped results, we divided the DCV
array into 3 parts and we tested the first and second part of DCV for each number (0 to9) Results
are shown in table (1).
We can get a unique value for the numbers (0, 3, 5, 6, 8) from the first part of the DCV array,
For numbers (1, 2, 4, 9) we need to test the second part of the DCV array. Once more, the first
part of DCV of the number 7 is not clear, thus, we need to test the second part to give the number
7 a unique value that can be used to recognize it.
From table (1) we can see that the DCV can show almost unique value of each number (0
to9) .We can use the tired part of the DCV to increase the recognition accuracy level. We can
divide the DCV into 6 parts and test each part for all numbers to increase the accuracy of the
system. That might be a part of our future work.
5. REFERENCES
1- Q. Wu, H. Zhang, W. Jia, X. He, J. Yang, T. Hintz (2006).Car Plate Detection Using
Cascaded Tree-Style Learner Based on Hybrid Object Features. Proceedings of the IEEE
International Conference on Video and Signal Based Surveillance (AVSS'06).
2- J.M. Lopez, J. Gonzalez, C. Galindo, and J. Cabello (2007). A versatile Low-cost Car Plate
Recognition System. Pan
3- Xiang Ye Xiuzi Zhang Sanyuan (2004). A Hybrid Method for Robust Car Plate Character
Recognition. IEEE International Conference on Systems, Man and Cybernetics.
4- S. Lu, Z. Liu, Y.Chen, and L. Liu (2008). AWHD for License Plate Character. The 2008
International Conference on Embedded Software and Systems Symposia (ICESS2008).
5- P Xiang, Y Xiuzi and Z Sanyuan (2004). A Hybrid Method for Robust Car Plate Character
Recognition. 2004 IEEE International Conference on Systems, Man and Cybernetics.
6- Qingchuan Tao. Xiaohai He, Daishen Luo, Wei Wu (2004) A New Car Plate Recognition
Method Based on Fuzzy Entropy Proceedings. World Congress on Intelligent Control and
Automation, June 15-19. 2004, Hangzhou. P.R. China.
7- W. Cheng and L. Shao-Fa (2008). An Adaptive Method of Car Plate Image Enhancement
Based on a Simplified Pulse Coupled Neural Network. International Symposium on
Computer Science and Computational Technology. International Conference on Machine
Learning and Cybernetics, Kunming, 12-15 July 2008.
8- Rafael C.Gonzalez, Richared E. Wood and Steven L.Eddins (2004). Digital Image
processing Using MATLAB. PEARSON. Prentice Hall, Upper Saddle River, NJ 07458
9- Wikipedia, the free encyclopedia (9-2-2009):
http://en.wikipedia.org/wiki/Automatic_number_plate_recognition
10- Wikipedia the free encyclopedia (9-2-2009):
http://en.wiktionary.org/wiki/Euclidean_distance
11- Wikipedia, the free encyclopedia (9-2-2009):
http://en.wikipedia.org/wiki/Fast_wavelet_transform
12- W. L. Chan, C. Hyeokho (2004). Directional hyper-complex wavelets for multidimensional
signal analysis and processing. IEEE International Conference on Acoustics, Speech,
and Signal Processing. Quebec,- Canada
13- H. C. Tan and H. Chen (2008). Novel Car Plate Verification with Adaptive Binarization
Method. Proceedings of the Seventh International Conference on Machine Learning and
Cybernetics, Kunming, 12-15 July 2008.
14- M. Misiti, Y. M. Georges O. Jean-Michel Poggi. Wavelet Toolbox User’ Guide. Copyright
1997-2008 by the MathWorks, Inc.
15- L Jin and J. Z. Qin (2003). Car Plate Number Character Recognition Using GABOR
Orientation Features and Neural Networks. IEEE Int. Conf. Neural Networks & Signal
Processing Nanjing, China, December 14-17,2003.
16- Y. Cheng, J Lu and T. Yahagi (2004). Car Licenses Plate Recognition Based on the
Combination of Principal Components Analysis Radial Basis Function Networks. 0-
7803-8406-7/04/IEEE.
17- R. Parisi, E.D.Di Claudio, G.Lucarelli and G.Orlandi(1998). Car Plate Recognition by
Neural Networks and Image Processing. 0-7803-4455-3/98/$10.000 1998 IEEE