Improved Method For License Plate Recognition For Indian Road Conditions

International Journal of Emerging Trends & Technology in Computer Science (IJETTCS)
Web Site: www.ijettcs.org Email: [email protected]

Volume 3, Issue 6, November-December 2014
ISSN 2278-6856
Improved Method for License Plate Recognition

for Indian Road Conditions
Poonam SKolhe1, Geeta Salunke2
1
G .S. Moze College of Engineering, Pune
All India Shri Shivaji Memorial Society's Institute of Information Technology
Abstract
The paper is presenting and evaluating the proposed methods
used in LPR system against existing methods in order to justify
the efficiency of proposed methods. These methods later may
be used for the applications like traffic control; vehicle owner
identification etc. This proposed approach split the LPR
applications into three major categories: access control (AC),
law enforcement (LE), and road patrol (RP). Each category is
characterized by variables of different variation scopes. This
proposed method consists of three modules for plate detection,
character segmentation, and recognition. .In this paper an
additional null class is introduced for improving character
recognition rate.In this papar, this proposed solution will
targeted the Indian datasets. In this paper the samples from
India, from various locations, time, traffic, and weather
condition. During the practical evaluation
measure the
performance of each module and compare it against the
existing technique in order to claim efficiency of proposed
solution
Keywords:-Character segmentation, plate detection,

vehicle license plate recognition (LPR).
1.
INTRODUCTION
During the past few years, As to improve transportation

safety and mobility and advanced technologies to increase
productivity through the use of intelligent transport
systems to their scope (ITSs) has had an extensive impact
in people's lives. ITSs technology based systems 16 types of
intelligent infrastructure systems and these systems.
Intelligent vehicle systems are split. in many ways recently
presented are based on the computer vision and a license
plate recognition (LPR) character recognition algorithm for
electronic Payment systems (pay toll, fee payment
parking), freeway, and to monitor arterial traffic
management systems like intelligent infrastructure is a key
to be used as presented. As increased safety awareness
vehicle arrived Edge authentication technologies have
made the need for extremely important, moreover, to enter
the private areas of the proposed system to unauthorized
vehicles for access control systems can be employed as
principal license plates. It may be deliberately altered or
fraudulent circumstances (for example, with a stolen plate)
that despite the fact the vehicle identifier. Therefore, ITSs
quite strong LPR systems rely on the focus of this paper a
novel split an LPR if properly parameterized external
circumstances capable of handling system implemented in
technology integration. Lack of frequent benchmark
database [6], an active research topic acknowledged
although LPR has been for more than a decade, [13]. Most
works proprietary data sets, which rarely are available for a
performance comparison report with evaluation on
performance. Several reviews cite only performance can in
fact it [6], [8], [12] without being able to validate reported.
The public lIe among those available, a limited number of
sampling or variable with a limited scope of [10]. A
suitable one Anagnostopoulos et al. [6], is given by the
variable scopes with different variation 741 images. It is,
however, not good enough to perform in the
aforementioned three major applications. In [1], recently
author license plate (AOLP) application-oriented
benchmark database, which is classified into three subsets
2049 images, presented the first version and each of the
three major applications sample subsets representing a
good scope. in various places, Taiwan time, traffic and
weather conditions all samples were collected is similar to
most LPR approaches.[1] in the solution presented plate
detection, character segmentation and recognition consists
of three modules, each module is in innovation. However
[1], the author took every step or modules offer insightful
analysis with attitude, and hence our proposed solution
could claim to efficiency.
2. LITERATURE SURVEY
There are many methods presented for LPR, discussing
them below according to their different phases.
2.1. Review of methods for Pre-processing
In [3], Otsu binarization method is used. The acquired
image get s segmented into several sub-regions. Threshold
value should be calculated for each sub regions.
Anagnostopoulos et.al. According to [4] pre-processing
ambient illumination conditions in the region of interest
(ROI) are used to detect it is image masking, using
binarization method with Sauvola. Sauvola method,
locally adaptive Thresholding is a binary image to gray
scale image is used to convert the value of the threshold
range. Like the surface of the fitting, variance and
parameters depends mainly on local stats. In the case of
badly illuminated areas is will reduce the calculated
threshold value. Chang et al, [5], for the purposes of
binarization mainly two: letter to highlight and to
suppress the background while some important
information will be lost from images binarization,
Thresholding a variable Nakagawa and employed the
technique proposed by Rosenfeld [5]. This technique, so
that the problem of non-uniform illumination avoid each
image pixel from a local optimum threshold value is
determined. Although local the adaptive Thresholding
Page 75

method completely offset the loss of information, it's at
least that may be lost when a constant binarization method
preserves the information using. In [6], a global threshold
value is chosen instead of an adaptive one. To minimize
the processing time, the original image isgetting down
sampled to 120 columns.G.Sun et al., in [7] divided the
pre-processing task into luminance adjustment and image
enhancement. T.Duan et.al. , [9], processing facilities to
enrich the performance edge of the level histogram
equalization algorithms are common and graying. Image
de-noising noise applied to remove images. This
procedure can be done in the broken link and tiny sudden
changes of image edges soft[11]. Outline and as much as
possible, increase ROI and other areas to maintain
contrast between lines, to prevent the destruction of this
technology is the main objective. [10],Bernsen, algorithm
effectively converts a binary image to remove the shadow
image from the original one.
2.2. Review of Methods for License Plate Detection
In [4], binary image successful CCL, orientation, aspect
ratio, Euler number count for each binary object in the
image are measurements. 35 , orientation, aspect ratio 2
<< 6 and Euler number >< like criteria 3 candidates as in
plate areas [4]. But this method that LP with dark
background and white letters will be found successfully is
not guaranteed. In [5] Fuzzy logic LP used to locate. To
explain about some rules and some membership functionsfor fuzzy set-"bright", "dark", "bright and dark sequence",
"texture" and "pilapa" of horizontal and vertical plate
positions, but to achieve this need is very sensitive to
colour and brightness LP it also compared to traditional
colour based methods no longer processing time. In [6]. P.
To determine the vertical edge detection method is used
then the Robert edge detector vertical edges are used .
There will happen sudden intensity changes, but a cluster
of 10-15 fast intensity changes in plate zone is regarded as
rank-filter . With M N pixels, convolved. This is a
bright spot in the area of ellipsoidal shape plate elongated.
The last step is horizontal projection.Sobel edge detection
in [11], apply the method after you apply certain rules to
determine the area of LP.
2.3. Character Recognition Methods
In [8], vertical and horizontal scanning is used to dig out
the characters. Vertical scanningl scan the image
vertically from [0, 0] to [height, width] .. In [2], different
methods are employed for character segmentation. At first
, a gray-level quantization and morphology analysis are
performed to obtain the candidate characters. In [10], For
characterrecognition,Support Vector Machine (SVM) is
widely used. Beforeusing SVM training and testing data,
features are extracted using two density methods i.e.LocalDirection Contributively Density (L-DCD) and GlobalDirection Contributively Density (G-DCD), SVM has less
misclassification rate compared to neural network. In [8],
template matching method in which the templates are
matched with existing character are used for character
recognition. Template matching stored with each
character blocks. Priority has been assigned to each
template is used when matching, the first top priority
ISSN 2278-6856
compared to the template and if a received the priorityis

lowest then matching template should be ignored .
Because chances are less than the number of alphabets
occurrences number templates are assigned high priorities.
In [6], there existstwo types of template matrix: Object
Thinned Representation (OTR) and Characteristics
Background Spots (CBS). OTR helps for representing the
character shape.
3. PROPOSED APPROACH FRAMEWORK AND
DESIGN
In this proposed method and its working is discussed. The
proposed system contains three approaches.
3.1. Problem Definition
The irregular shape of the shade and the vehicle license
plate characters some factors, like a font and background
color affect the performance of some LPR systems work
only in restricted terms and adversity may not produce
good amount of accuracy. systems are developed and some
of the specific country are there used to poll very few
systems are for India but it has been observed that Those
people are not exercising more accurate then to develop a
system for a country like India a comprehensive scope. In
this paper the main aim is to present the extended method
for LPR especially for Indian conditions with aim of
improving the overall accuracy. Apart from the above
discussed below are the objectives of this project:
To focus on review of different methods presented for
LRP.To focus on the design of proposed methods,
datasets..
3.2. Proposed Architecture and methodology
Following figure 1 showing the proposed project
methodology and framework:
Page 76

Figure 1: Flowchart of the overall proposed system with

three modules for plate detection, character segmentation,
and recognition.
The proposed system contains image collection from
various Indian datasets. Preprocessing can be done by
using various methods. Edge detection carried out using
sobeledge detector.LPdetcction and extraction should be
done by GMM and EM clustering.Character segmentation
is done by MSER method and character recognition can be
done by OCR methods.
ISSN 2278-6856
Fig 3:Input image and preprocessed image

4.4. Detection of plate candidate:
The output of the preprocessed image given to sobel edge
detector .Edge detection is the process of localizing pixel
intensity transitions. There exists several edge detection
methods Prewitt, Sobel,Robert.From this we are using
Sobel edge detection for its simplicity and easy to use .
Sobel edge detector uses two masks one horizontal and
other is vertical.Sobel edge detection requires grey scale
image.
4. SPECIFICATIONS
This paragraph contains the specification of hardware
and software.
4.1. Input:
The Input Image for Processing License Plate Detection
capture from Indian Dataset.
4.2. Hardware and Software Configuration
Hardware Specification:
Processor
: Pentium IV 2.6 GHz
Ram : 512 MB DD RAM
Monitor
: 15 COLOR
Hard Disk
: 20 GB
Software Specification:
Front End
: Matlab
Tools Used
: Matlab 2012
Operating System : Windows 7/8
4.3. Pre-processing
PreProcessing includes converting the image into grey
scale ,image is converted into grey scale by setting up the
threshold and image sharpening and smoothingoperation
is carried out. The converted image is converted into
binary image and preprocessed image is provided to edge
detection algorithm for locating the plate.
Fig4:Edges extracted by Sobel edge detector.

The output of the sobel edge detector is given to GMM
and EM clustering algorithm and output is detected as a
plate candidates. Clustering can be used because of its
extensive coverage of different scales, orientations, and
rotations in one session. The output of clustering is look
like as follows.we are increasing the clusters one by one
and selecting the most appropriate output.The clusters
which are accepted by size,
shape,density and
orientation.The clusters which are staying Stationary even
with the number of clusters are increasing.
Fig 2: Preprocessing flow chart of vehicle image
Page 77

ISSN 2278-6856
scheme is to analyze the MSER+ and MSER- detection

results. We are looking for a larger MSER+region (license
plate) that contains a set of smaller MSER- regions
(characters).Such a combination is considered as license
plate detection result. After verification, the MSER+ is
returned as license plate localization and additionally
segmentations of the characters are provided by the
corresponding MSER- detections.
Fig6:Extracted MSER features.
Fig5:Objects in cluster1 and Objects in Cluster2 and

Objects in cluster 3
4.5. CharacterSegmentation:
A plate candidate from the detection module, is given to
an MSER detector forexploiting possible characters. The
MSER
was
originally
proposed
for
finding
correspondences across viewpoints .It can render the
persistentedges around objects as illumination changes, it
was proven to be effective in segmenting the characters
that often reveal edges robust to illumination
variation.The detection algorithm is based on analysing
the results of Maximally Stable External Region (MSER)
detection [8]. MSERs denote a set of distinguished regions
and have proven to be one of the best interest point
detectors in computer vision [9]. All of these regions are
defined by an external property of the intensity function in
the region and on its boundary, MSERs are detected in
every scale. We predominantly exploit these properties for
segmentation purposes. In general, two variants of MSER
detection can be distinguished denoted as MSER+ and
MSER-. While MSER+ detects bright regions with darker
boundary, MSER- finds dark regions with brighter
boundary . The license plate itself is identified as MSER+,
whereas the characters on the plate are detected as MSER1. MSER detection results can be used for detecting
license plates in video sequences. MSER+ finds the
license plate, whereas MSER- identifies the characters on
it. The underlying idea of our novel license plate detection
4.6. Character Recognition

The optical character recognition (OCR) is a recognition
method in which the input is an image and the output
gives us a string of character. OCR is a process which
separates theindividual characters from image. Template
matching is one of the approaches of OCR. The cropped
image is compared with the template stored in database
.OCR automatically identifies and recognizes the
characters. The characters on the number plate have their
uniform fonts and sizes then the OCR for number plate
recognition is less complex as compared to other methods.
After Preprocessing of characters we have to do the
template matching so that we can match the individual
score of the letters and identify the numbers and letters on
license plate. With the inbuilt imcrop() function of matlab
we can crop the image and the result is written in a text
file.
Fig7:Bounded Box Cropping
Fig8: Detected License Plate
5. CONCLUSION AND FUTURE WORK:

The proposed approach is divided into three major
categories: LPR applications AC, traffic LE and RP. Each
category is characterized by variable scopes different
variation. The results of edge detection is satisfactory but
the time required of it is high it is more than 14 sec and
secondly if the image quality is not good that is when the
Page 78

ISSN 2278-6856
the focus varies we will not get very accurate results.so the
algorithms should be improved. The MSER is proven
effective in character segmentation in this paper, and
otherinterest region detectors deserve to be studied for
their potentialin improving LPR performance.
REFERENCES
[1] C. Anagnostopoulos, I. Anagnostopoulos, V. Loumos,
and E. Kayafas, A license plate-recognition algorithm
for intelligent transportation system applications,
IEEE Trans. Intell. Transp. Syst., vol. 7, no. 3, pp.
377392, Sep. 2006.
[2] J.-M. Guo and Y.-F.Liu, License plate localization
and character segmentation with feedback selflearning and hybrid binarization techniques,IEEE
Trans. Veh. Tech., vol. 57, no. 3, pp. 14171424, May
2008.
[3] K.
Suresh,
G.
Kumar,
and
A.
RajagopalanSuperresolution of license plates in real
traffic videos, IEEE Trans. Intell. Transp. Syst., vol.
8, no. 2, pp. 321331, Jun. 2007.
[4] S.-L. Chang, L.-S.Chen, Y.-C.Chung, and S.-W.
Chen, Automatic license plate recognition, IEEE
Trans. Intell. Transp. Syst., vol. 5, no. 1, pp. 4253,
Mar. 2004.
[5] H. Caner, H. Gecim, and A. Alkar, Efficient
embedded
neural-networkbased
license
plate
recognition system, IEEE Trans. Veh. Tech., vol. 57,
no. 5, pp. 26752683, Sep. 2008.
[6] C.-N. Anagnostopoulos, I. Anagnostopoulos, I.
Psoroulas, V. Loumos, and E. Kayafas, License plate
recognition from still images and video sequences: A
survey, IEEE Trans. Intell. Transp. Syst., vol. 9, no.
3, pp. 377391, Sep. 2008.
[7] H. Sheng, C. Li, Q. Wen, and Z. Xiong, Real-time
anti-interference location of vehicle license plates
using high-definition video, IEEE Intell.Transp. Syst.
Mag., vol. 1, no. 4, pp. 1723, 2009.
[8] J. Jiao, Q. Ye, and Q. Huang, A configurable method
for multi-style license plate recognition, Patt.Recog.,
vol. 42, pp. 358369, Mar. 2009.
[9] X. Fan and G. Fan, Graphical models for joint
segmentation and recognition of license plate
characters, IEEE Signal Process. Lett., vol. 16, no. 1,
pp. 1013, Jan. 2009.
[10] H. Tuna, I. Onaran, and A. Cetin, Image description
using a multiplierless operator, IEEE Signal Process.
Lett., vol. 16, no. 9, pp. 751753, Sep. 2009.
[11] Z.-X. Chen, C.-Y.Liu, F.-L.Chang, and G.-Y. Wang,
Automatic licenseplatelocation and recognition based
on feature salience, IEEE Trans. Veh. Technol., vol.
58, no. 7, pp. 37813785, Sep. 2009.
[12] Y. Wen, Y. Lu, J. Yan, Z. Zhou, K. von Deneen, and
P. Shi, An algorithm for license plate recognition
applied to intelligent transportation system,IEEE
Trans. Intell. Transp. Syst., vol. 12, no. 3, pp. 830
845, Mar. 2011.
Page 79

Improved Method For License Plate Recognition For Indian Road Conditions

Uploaded by

Copyright:

Available Formats

Improved Method For License Plate Recognition For Indian Road Conditions

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Improved Method For License Plate Recognition For Indian Road Conditions

Uploaded by

Copyright:

Available Formats

International Journal of Emerging Trends & Technology in Computer Science (IJETTCS)

Web Site: www.ijettcs.org Email: [email protected]