a review: recognition of automatic license plate in … · a review: recognition of automatic...

International Journal of All Research Education and Scientific Methods (IJARESM)

ISSN: 2455-6211, Volume 4, Issue 6, June- 2016

92

A Review: Recognition of Automatic License

Plate in Image Processing

Babita1, Ms. Meenu

2

1M.Tech Student, CSE Deptt, VCE Rohtak

2Assistant Professor, CSE Deptt, VCE Rohtak

ABSTRACT

In this paper a detailed study on Automated License plate recognition is given. In this paper we are proposing a

method for implementing Automatic License Plate Recognition using Speed up robust feature matching SURF

technique for plate detection and Advanced Radial basis function for matching characters. Automatic License Plate

Recognition algorithms generally consist of following steps for effective processing: - 1) Identification and detection of

the license plate area; 2) Segmentation of the plate characters; and 3) Optical Character Reader. Image processing

techniques such as filtering, edge detection, Thresholding etc are normally used to perform the first two steps of the

ALPR process. This proposed work has been implemented by image processing Toolbox under the MATLAB

software.

Keywords: SURF, RBF, Neural Network, OCR, MATLAB, Segmentation, Radial Basis Function and Ransac.

1. INTRODUCTION

Computer vision techniques have led to new innovation in the automation in license plate localization. Number

plate localization has been a major and vital part of the license plate recognition system .This technique can

solve various traffic management problems such as Highway toll collection, Parking management, vehicle Theft

detection etc [3].

2. LITERATURE REVIEW

Character Recognition is a technique which is most widely used for authentication of person as well as

document. Character Recognition is performed On-line as well as Off-line. Optical Character Recognition is

performed Off-line after the writing has been completed [2]. In Off-line, each character is differentiating by

analysis of character shape and comparing the features . [5]. Whereas, In On-line Recognition, computer

recognizes the character as they are drawn [2].

In Character Recognition System, firstly a database is prepared which consists of printed English alphabet. Then,

the characters are scanned and pre-processing technique is used for the detection and removal of noise and

improve the quality of images. Various methods used in character recognition are Image Pre-processing,

Segmentation, Feature Extraction, Classification and Recognition and Pattern Matching [1,3]. Techniques used

in this are Neural Network Algorithm, Image processing Toolbox created in MATLAB software, SURF Feature,

Ransac Algorithm and Radial Basis Function [4,6].

3. PROPOSED METHOD

A. Optical Character Recognition

Optical Character Recognition (OCR) using Neural Netwok is basically used in the field of research. Character recognition technique is mainly used for tolerate the defects in license plate such as license plate may be tilted or

bent with respect to position of camera. The proposed method consists of basically topological sorting method

which includes the number of end points, number of holes. This approach provides the feature in character

recognition with quality despite of quantitative nature. Steps of Optical Character Recognition are shown in fig

1.

B. Steps of Optical Character Recognition



93

Fig 1: Steps of Optical Character Recognition [4]

1. Image Acquisition

In Image Acquisition, this approach acquires an input in the form of scanned image and also provides the

location from where it is acquired. All the scanned images which we are using must be in a specific format such

as JPEG, JPG etc. These images can be input to the system through any type of suitable input device such as scanner, digital camera or any analog or digital input device. Image Acquisition process is shown in fig.2 [4].

Fig 2: Sample Dataset [2]

2. Pre-Processing

The pre-processing consist a series of steps which are performed on the raw data to prepare it for another

processing . It basically consists of three tasks labelling of connected components, thresholding and removal of

noise which are performed in a sequential manner. The labelling of components selects the useful data from a

large collection of data Normally, in this approach, transformation is used to manipulate raw data to produce a

single input and normalization is used to manage the data in a more suitable way so that ease to access. The

process of Pre-processing is shown in fig 3. [1]

Fig 3: Gray-Scale Image [1]

i) RGB to Gray Conversion: First of all the coloured image is converted into a gray colour format and the

scanned image is stored in a JPEG format but we can also use the other formats of images like BMP, JPG etc for character recognition. There is a specified format for these image called RGB format. A converted gray scale

image compromises of number of pixels which are expressed within a minimum and maximum value and lies

between a range of 0 to 255 where 0 shows the total absence i.e. black and 1 shows the total presence i.e. white.



94

In this process each pixel contains a single set called black and white and composed of shades of gray where

black at the weakest intensity and white at the strongest. [5].

ii) Thresholding Approach or Binarization: Thresholding is a process in which a image consisting of Gray

Scale is converted into a image of 0 and 1 i.e. machine readable form Image using a technique named as

Thresholding. This approach also stores an image in the form of row and column known as matrix form , but in this a point consisting of pixels can only be coloured as either Black or White. It assigns value True i.e. 1 for

white and False i.e. 0 for Black.

iii) Noise Reduction: connected line segments can be affected by Noise results to disconnection of line

segments, bumps, and also creates gaps in lines and filled loops which are created by either scanning device or

any other suitable writing approach. A process, named as Median filter is used for reducing the noise. By using

Filter process of Median , the value of a point consisting of pixels is replaces by the median of gray levels near

of that pixel [1].

iv) Thinning: This approach is morphological operation process. It is used in binary image to remove the

foreground pixels from the scanned images. This process takes the archaeological image as an input and after

performing the thinning process it provides an thinned image of one pixel width characters. This approach can be done by using two types of algorithms: Sequential and parallel Thinning algorithms.

[2].

v) Edge Detection Approach and Erosion: Edge detection process is used to find the boundaries of objects

within images. To produce the pre-processed image for segmentation, these operations are performed in the last

two stages. Edge detection consists of mainly three steps:-

a. Smoothing

b. Edge enhancement

c. Edge localization

3. Detection/Localisation

i) SURF: For Licence Plate Localization, a method is used named as SURF feature extraction. SURF method

which is used for feature detection and extraction is based on Hessian Matrix. SURF method is basically

dependent on the determinant of Hessian Matrix. For Hessian Matrix, we use a continuous function of two

variables with the value of the function at (x; y) is given by f(x; y). Hessian matrix, H, is denoted as the matrix

of partial derivates of the function f . SURF Feature Extraction method is shown in fig 4. [3].

Fig 4: Matched features [3]

ii) Ransac: Ransac stands for Random Sample Consensus. Ransac algorithm is used for finding outlier

detection after finding the interest point for accurate detection. For finding outliers we provided parameters of

inliers and outliers based on which we removed the regions that were wrongly classified as plate region.

The RANSAC algorithm is consists of basically two steps which are performed in repeated manner,

Hypothesize and Test framework [3].

(a) Hypothesize: From set of input datasets minimal sample sets (MSSs) are selected firstly. It is a formal

statement of relationship between two or more variables. The instances are considered using only minimal

sample sets (MSSs) which consist of a number of elements It provides the facility of data collection, data

analysis and data interpretation.[3].



95

(b) Testing of Framework: In the first step, by using the model parameters, RANSAC approach is used to

check the elements in dataset for removal of inconsistency. The collection of these elements is known as

Consensus set (CS). RANSAC approach will quit only when the probability of finding a batter ranked ConS

will be less than a specific value. The number plate with inliers features is shown in fig 5 [3]

Fig 5: Located number plate with inliers features [3]

4. Segmentation

Segmentation is a process in which we segments the latters in different parts and stores them at some folder.

Morphological operations are used for segmentation. In this dilation, close, open operations are used to find the

connected components lower than a certain threshold value and extracted all the components as characters.

These characters are later fed into recognition module of ANN [7].

Some of the segmentation results are shown in fig: 6.

Fig 6: Segmented letters in GUI [7]

5. Pattern Matching

Pattern Matching comes when the segmentation process is complete and each character has been individually

extracted. In Pattern Matching, we created a database of 26 alphabets and numbers of size 24x42 pixels using

Adobe Photoshop CS3. This database contains multiple samples of one character for Neural Network training

purpose. Using Pattern Matching step, the speed and accuracy of overall recognition process is increased. For

Pattern Matching, RBF network is used. RBF network is a Radial Basis Feed Forward Propagation Neural



96

Network which consist of basically three layers: Input, Hidden and Output. Basic structure of RBF is shown in

fig7 [2].

Fig 7: RBF basic structure [2]

In this figure there are a number of inputs from p1,p2 to pr and input to the radial basis transfer function(radbas)

is a vector which is used to show the distance between weight vector w and the input vector p, multiplied by the

bias b. (The || dist || box in the above fig. The radial bias function accepts p as the input vector and the single

row input weight matrix, and produces the output as a dot product of the two.) In RBF network, we use the input

layer for providing the signal and then all the weights are connected to hidden layer which have assigned a value

one and radial basis function is used as an activation function [2].

4. TECHNIQUES USED

NEURAL NETWORK: A Neural Network can be thought as a collection of number of nodes which are

interconnected to each other which provides an adaptive neural processor. Firstly, the basic reason behind its

higher speed is that it works on the procedure of parallelism; it can perform the calculation at a higher rate as

compared to the classical techniques. Because a neural network provides a dynamic environment, so it provides

ease to changes in data and learn the features of input signal which are provided by input layer. In this network

output from one node is fed to another node and the final output resides on the complex co-operations of all

nodes. In this architecture output of one node fed in to input of other nodes and follows in a sequential manner.

The structure of Neural Network is shown in fig 8. [2]

Fig 8: Neural Network [2]

Characteristics of Neural Network

a. Training and Learning

b. Fault Tolerant



97

c. Evidential Response

d. Adaptively

e. Activation Function

f. Generalization

g. Noise Immunity

CONCLUSION

The main aim of this paper is to simulate and examine the results of Automatic License Plate Recognition using

Speed Up Robust Feature (SURF) Method for detection of License Plate Region. In this we have also used

Radial Basis Function (RBF) for character recognition which is successful in recognizing characters segmented

with Blurry and Noisy background. All the experiments and simulations have been performed with MATLAB

2011.

REFRENCES

[1]. Shradha Chadokar and Jijo S Nair. “A Survey on Offline Prompt Signature Recognition and Verification”.

International Journal of Advanced Research in Computer Science and Software Engineering, Vol. 4, pp. 1432-1438 (2014).

[2]. Reetika Verma and Mrs. Rupinder kaur. ” Enhanced Character Recognition Using SURF Features and Neural Network Techniques”.

[3]. International Journal of Computer Science and Information Technologies. Vol. 5, pp. 5565-5570(2014). [4]. Abhay Chaturvedi and Nidhi Sethi. “Automatic Licence Plate Recognition System SURF Features and RBF

Neural Network”. International Journal of Computer Application, Vol. 70, pp. 37-41 (2013).

[5]. Kauleshwar Prasad, devvrat C. Nigam, Ashmika Lakhotiya and Dheeren Umre.” Character Recognition Using Matlab’s Neural Network Toolbox “.International Journal of U- and e- service, Science and Technology. Vol. 6, pp. 13-20(2013).

[6]. Md. Alamgir Badsha, Md. Akkas Ali, Dr. Kaushik Deb and Md. Nuruzzaman Bhuiyan. “Handwritten Bangla Character Recognition Using Neural Network”. International Journal of Advanced Research in Computer Science and Software Engineering. Vol. 2, pp. 307-312(2012).

[7]. Pal S, Chanda S, Pal U, Franke K and Blumenstein M. “Off-line Signature Verification Using G-SURF”. Intelligent System Design and Applications, Vol. 2, pp. 586-591(2012).

[8]. Gaganjot Kaur, Monika Agarwal. “Artificial Intelligent System for Character Recognition Using Levenberg-Marquardt Algorithm”. International Journal of Advanced Research in Computer Science and Software Engineering, Vol. 2, Issue 5, pp. 220-223(2012).

[9]. Amit Choudhary and Rahul Rishi. “Improving the Character Recognition Efficiency of Feed Forward Neural Network”. International General of Computer Science and Information Technology, Vol. 1, pp. 85-96(2011).

[10]. Patel C.I., Patel R. and Patel P. “Handwritten Character Recognition Using Neural Network”. International Journal of Computer Science and Engineering Research, Vol. 2, pp. 1-6(2011).

[11]. Anita Pal and Dayashankar Singh. ” Handwritten English Character Recognition Using Neural Network.”

International Journal of Computer Science and Communication, Vol. 1, pp. 141-144(2010). [12]. Bansal R, Sehgal P and Bedi P. “ Effective Morphological Extraction of True Finger Print Minutial Based on the

Hit and Miss Transform.” International Journal of Biometrics and Bioinformatics, Vol. 4, pp. 71-85(2010).

a review: recognition of automatic license plate in … · a review: recognition of automatic...

Documents