Automatic Number Plate Recognition System (ANPR) A Survey.pdf

International Journal of Computer Applications (0975 – 8887)
Volume 69– No.9, May 2013
21
Automatic Number Plate Recognition System
(ANPR): A Survey 
Chirag Patel
Smt. Chandaben Mohanbhai
Patel Institute of Computer
Applications (CMPICA)
Charotar University of Science
and Technology(CHARUSAT)
Dipti Shah, PhD.
G. H. Patel P.G. Department of
Computer Science
S. P. University
Atul Patel, PhD.
Smt. Chandaben Mohanbhai
Patel Institute of Computer
Applications (CMPICA)
Charotar University of Science
and Technology(CHARUSAT)
ABSTRACT
Traffic control and vehicle owner identification has become
major problem in every country. Sometimes it becomes
difficult to identify vehicle owner who violates traffic rules
and drives too fast. Therefore, it is not possible to catch and
punish those kinds of people because the traffic personal
might not be able to retrieve vehicle number from the moving
vehicle because of the speed of the vehicle. Therefore, there is
a need to develop Automatic Number Plate Recognition
(ANPR) system as a one of the solutions to this problem.
There are numerous ANPR systems available today. These
systems are based on different methodologies but still it is
really challenging task as some of the factors like high speed
of vehicle, non-uniform vehicle number plate, language of
vehicle number and different lighting conditions can affect a
lot in the overall recognition rate. Most of the systems work
under these limitations. In this paper, different approaches of
ANPR are discussed by considering image size, success rate
and processing time as parameters. Towards the end of this
paper, an extension to ANPR is suggested.
Keywords
Automatic Number Plate Recognition (ANPR), Artificial
Neural Network (ANN), Character Segmentation Image
Segmentation, Number Plate, Optical Character Recognition
1. INTRODUCTION
1.1 Automatic Number Plate
Recognition (ANPR)
In last few years, ANPR or license plate recognition (LPR)
has been one of the useful approaches for vehicle surveillance.
It is can be applied at number of public places for fulfilling
some of the purposes like traffic safety enforcement,
automatic toll text collection [1], car park system [2] and
Automatic vehicle parking system [3]. ANPR algorithms are
generally divided in four steps: (1) Vehicle image capture (2)
Number plate detection (3) Character segmentation and (4)
Character recognition. As it is shown in Fig.1, the first step
i.e. to capture image of vehicle looks very easy but it is quite
exigent task as it is very difficult to capture image of moving
vehicle in real time in such a manner that none of the
component of vehicle especially the vehicle number plate
should be missed. Presently number plate detection and
recognition processing time is less than 50 ms [4] in many
systems. The success of fourth step depends on how second
and third step are able to locate vehicle number plate and
separate each character. These systems follow different
approaches to locate vehicle number plate from vehicle and
then to extract vehicle number from that image. Most of the
ANPR systems are based on common approaches like
artificial neural network (ANN) [5], [1],[6],[7][8], [9], [10],
Probabilistic neural network (PNN) [11], Optical Character
Recognition (OCR)[5], [12], [2], [13], [7], [14], Feature
salient [15], MATLAB[16], Configurable method[17], Sliding
concentrating window (SCW)[14],[8], BP neural network[18],
support vector machine(SVM)[19], inductive learning [20],
region based [21], color segmentation [22], fuzzy based
algorithm [23], scale invariant feature transform (SIFT) [24],
trichromatic imaging, Least Square Method(LSM) [25],[26],
online license plate matching based on weighted edit distance
[27] and color-discrete characteristics [28]. A case study of
license plate reader (LPR) is well explained in [29]. Some
authors focus on improving resolution of the low-resolution
image by using technique called super resolution [30], [31].
Sometimes it becomes necessary to assess the quality of
ANPR system. In [9], a quality assessment of visual and
ANPR is well explained. A comprehensive study of License
plate recognition (LPR) is well presented in [4]. Throughout
this literature, number plate and license plate are used
interchangeably. Details about each ANPR are discussed in
section 2.
Fig.1. Conventional ANPR system
.

22
1.2 Scope of this paper
As it is not possible to judge which approach is better,
different papers, which are based on steps mentioned in Fig.1,
are surveyed and categorized based on the methodologies in
each approach. For each approach whenever available
parameters like speed, accuracy, performance, image size and
platform are reported. Commercial product survey is beyond
the scope of this paper as sometimes these products claims
more accuracy than actual for promotional purposes.
Remainder of this paper is divided as follows: Section 2
contains survey of different techniques to detect number plate.
Character segmentation methods are reviewed in section 3 and
section 4 contains discussion about character recognition
methods. The paper concludes with the discussion of what is
not implemented and what kind of research is possible in
ANPR.
2. NUMBER PLATE DETECTION
Most of the number plate detection algorithms fall in more
than one category based on different techniques. To detect
vehicle number plate following factors should be considered:
(1). Plate size: a plate can be of different size in a vehicle
image.
(2). Plate location: a plate can be located anywhere in the
vehicle.
(3). Plate background: A plate can have different background
colors based on vehicle type. For example a government
vehicle number plate might have different background than
other public vehicles.
(4). Screw: A plate may have screw and that could be
considered as a character.
A number plate can be extracted by using image segmentation
method. There are numerous image segmentation methods
available in various literatures. In most of the methods image
binarization is used. Some authors use Otsu’s method for
image binarization to convert color image to gray scale image.
Some plate segmentation algorithms are based on color
segmentation. A study of license plate location based on color
segmentation is discussed in [22]. In the following sections
common number plate extraction methods are explained,
which is followed by detailed discussion of image
segmentation techniques adopted in various literature of
ANPR or LPR.
2.1 Image binarization
Image binarization is a process to convert an image to black
and white. In this method, certain threshold is chosen to
classify certain pixels as black and certain pixels as white. But
the main problem is how to choose correct threshold value for
particular image. Sometimes it becomes very difficult or
impossible to select optimal threshold value. Adaptive
Thresholding can be used to overcome this problem. A
threshold can be selected by user manually or it can be
selected by an algorithm automatically which is known as
automatic thresholding.
2.2 Edge detection
Edge detection is fundamental method for feature detection or
feature extraction. In general case the result of applying edge
detection of algorithm is an object boundary with connected
curves. It becomes very difficult to apply this method to
complex images as it might result with object boundary with
not connected curves. Different edge detection algorithm /
operators such as Canny, Canny-Deriche, Differential, Sobel,
Prewitt and Roberts Cross are used for edge detection.
2.3 Hough Transform
It is a feature extraction technique initially used for line
detection. Later on it has been extended to find position of
arbitrary shape like circle or oval. The original algorithm was
generalized by D.H. Ballard [32].
2.4 Blob detection
Blob detection is used to detect points or regions that differ in
brightness or color as compared to surroundings. The main
purpose of using this approach is to find complimentary
regions which are not detected by edge detection or corner
detection algorithms. Some common blob detectors are
Laplacian of Gaussian (LoG), Difference of Gaussians (DoG),
Determinant of Hessian (DoH), maximally stable extremal
regions and Principle curvature based region detector.
2.5 Connected Component Analysis
(CCA)
CCA or blob extraction is an approach to uniquely label
subsets of connected components based on a given heuristic.
It scans binary image and labels pixel as per connectivity
conditions of current pixel such as North-East, North, North-
West and West of the current pixel (8-connectivity). 4-
connectivity is used for only north and west neighbour of
current pixel. The algorithm gives better performance and it is
very useful for automated image analysis. This method can be
used in plate segmentation as well as character segmentation.
2.6 Mathematical morphology
Mathematical morphology is based on set theory, lattice
theory, topology, and random functions. It is commonly
applicable to digital image but can be used in other spatial
structures also. Initially it was developed for processing
binary images and then extended for processing gray scale
functions and images. It contains basic operators such as
Erosion, dilation, opening, closing.
2.7 Related work in number plate
detection
The methods discussed in preceding sections are
common methods for plate detection. Apart from these
methods, various literature discussed method for plate
detection. As most of the methods discussed in these
literatures use more than one approach, it is not possible to do
category wise discussion. Different number plate
segmentation algorithms are discussed below.
In [5], for faster detection of region of interest
(ROI) a technique called sliding concentric window (SCW) is
developed. It is a two step method contains two concentric
windows moving from upper left corner of the image. Then
statistical measurements in both windows were calculated
based on the segmentation rule which says that if the ratio of
the mean or median in the two windows exceeds a threshold,
which is set by the, then the central pixel of the windows is
considered to belong to an ROI. The two windows stop
sliding after the whole image is scanned. The threshold value
can be decided based on trial and error basis. The connected
component analysis is also used to have overall success rate of
96%. The experiment was carried out on Pentium IV at 3.0

23
GHz with 512-MB RAM and took 111ms of processing time
for number plate segmentation.
Another SCW based system is presented in [8] for
locating Korean number plate. After applying SCW on
vehicle image authors used HSI color model for color
verification and then tilt was corrected by using least square
fitting with perpendicular offsets (LSFPO). The distance
between camera and vehicle varies from 3 to 7 meters.
A cascade framework was used in [33] for
developing fast algorithm for real time vehicle number plate
detection. In this framework a compact frame detection
module is used to segment number plate. This module
contains three steps: First - Generation of Plate Region
Candidates which is used to reject non plate regions by using
gradient features. Second – Extraction of complex plate
regions which contains three steps to identify plate region and
reject non plate regions. Third – plate verification is used to
make sure that no non plate regions are extracted in preceding
steps. The experiment was carried out on 3-GHz Intel Pentium
4 personal computer.
To detect multi-style number plate a configurable
method is proposed in [17]. For detecting different style of
number plates, a user can configure the algorithm by changing
parameter value in the number plate detection algorithm. The
authors define four parameters mainly:
 Plate rotation angle- to rotate number at certain angle
plate if it is skewed which is shown in fig. 2(a).
 Character line number – to determine whether
characters are spanned in more than one line or
column as shown in fig. 2(b). The algorithm works for
maximum three lines.
 Recognition models – to determine whether number
plate contains alphabets only, alphabets and digits or
alphabet, digits and symbols.
 Character formats – To classify the number plate
characters based on their type. For example, Symbols
can be represented as S, Alphabets can be represented
as A and digits can be represented as D. So the number
in fig.2, can be represented as AADAADDDD.
The algorithm was executed on Pentium IV 3.0GHz.
To locate Indian number plate, a feature based
number plate localization is proposed in [34]. The authors use
Otsu’s method to convert gray scale images into binary
images. It is a seven-step procedure to extract number plate
without any background image from vehicle image.
In [15] a feature salient method is used to extract
vehicle number plate by using salient features like shape,
texture and color. The authors used Hough transform (HT) to
detect vertical and horizontal lines from rectangular vehicle
number plate and then processed it by converting red, green,
blue (RGB) to hue-intensity-saturation(HIS). Finally, the
number plate is segmented. This algorithm is executed on
Pentium-IV 2.26-GHz PC with 1 GB RAM using MATLAB.
An Improved bernsen algorithm is used in [19] for
license plate location. This algorithm is used for the
conditions like uneven illumination and particularly for
shadow removal. The authors used local Otsu, global Otsu,
and differential local threshold binary methods for good
accuracy. By using this algorithm, shadow was removed and
license plate was successfully detected, which was not
possible with the traditional bernsen algorithm. The
experiment was carried out on Windows XP Operating system
Intel Core 1.8 GHz central processing unit and 1.5 GB RAM.
The algorithm was developed using Visual C++.
To locate Chinese number plate Hui Wu and Bing
Li [35] proposed a method to find horizontal and vertical
difference to find exact rectangle with vehicle number. The
Authors converted vehicle image into gray scale and then
applied automatic binarization using MATLAB. Any further
detail regarding number plate detection algorithm is not
mentioned in this paper. The authors claim to have average
recognition rate of 0.8s.
To extract license plate characters in Indian
condition Ch.Jaya Lakshmi et al. [16] proposed a novel
approach which is based on texture characteristics and
wavelets [36]. The authors also used morphological operation
[37] for better performance in complicated background. Sobel
mask is used to detect vertical edges. The system was
implemented using MATLAB. A Sobel edge detector operator
is also used in [38].
To detect license plate from CCTV footage,
M.S.Sarfraz et al.[39] proposed a novel approach for efficient
localization of license plates in video sequence and the use of
a revised version of an existing technique for tracking and
recognition. The authors proposed a novel solution for
adjusting varying camera distance and diverse lighting
conditions. License plate detection is a four step procedure
including finding contours and connected components,
selection of rectangle region based on size and aspect ratio,
initial learning for adaptive camera distance/height,
localization based on histogram, gradient processing, and
nearest mean classifier. After processing these steps final
detection result is forwarded for tracking.
In [6], canny edge detector operator was applied
to find out the transition points. As per H.Erdinc Kocer et al a
license plate contains white background and black character
normally. The Canny edge detector uses a filter, which is then
based on Gaussian smoothing’s first derivative to eliminate
the noise. Then in the next step, the edge strength is calculated
by considering the gradient of the image. The canny edge
detector operator used 3 X 3 matrix to accomplish this task.
Based on this information transition points region is
determined. The edge map is used to find transition points
between black and white colors. The further technical details
of this algorithm are not mentioned. The vehicle images were
captures from CCD camera.
(a) Skewed image (b) Number plate with lines
Fig. 2. Vehicle number plate with first two parameters
as per [17]

24
For detecting number plates of different countries
Ankush Roy et al. [7] presented improved segmentation
algorithm. The number plate segmentation algorithm is a four
step procedure including median filtering, adaptive
thresholding, component labeling and region growing and
segmentation and normalization to remove noise, for
binarization of image, to label the pixel according to color
value and to segment the plate of 15 X 15 pixel size. The
authors used Otsu’s method for image binarization in the
adaptive thresholding process. The overall success rate of
system is mentioned but success rate of number plate
detection rate is not mentioned in this paper.
In [14], global edge features and local Haar-like
features are proposed for real-time traffic video. License plate
detection is accomplished by moving a scanning window
around the vehicle image. The scanning windows is
categorized a license plate region and non license plate region
based on the pre-defined classifier. In the training phase, six
cascade classifier layers are constructed for future processing.
In the testing phase, local Haar-like features and global
features are extracted. Haar-like features are the digital image
features generally used for object segmentation. These
features are generally collection of functions to find number
of rectangles covering adjacent image regions. Global features
include edge density and edge density variable. These features
are calculated by using fix size of sample image i.e. 48 X 16
which is scaled in training phase. The experiment was carried
out on a PC with Pentium 2.8 GHz CPU. The average
processing time for segmentation was 0.204s. Another edge
based number plate segmentation algorithm is presented in
[38].
A weighted statistical method is applied in [18].
Before processing further, authors converted 24 bit color
image into gray scale image. In the weighted statistical
method, a 2D image matrix of N rows and M columns is
prepared. Then weighting operation is applied to the modified
image matrix prepared after adding weights. As per Zhigang
Zhanga et al. standard license plate length and breadth
proportion is 3.14:1. More implementation details regarding
this method are not mentioned in this paper. Similar method is
proposed in [40].
To detect license plate in varying illumination
conditions a novel approach is presented in [41]. A
binarization method is used as pre-processing step for plate
segmentation. In this approach, the author divided small
window region in the image and applied dynamic thresholding
method to each region. As per Naito, T et al. [41], this method
is very robust for a local change in brightness of an image.
Then binarized image is labelled and segmented as per the
algorithm mentioned in this paper. The algorithm works for
Japanese number plate but it is not clear that whether it can be
applied for other countries. The experiment environment and
processing time are not mentioned.
In [42] a novel approach based on vector
quantization (VQ) is discussed. As per Zunino, R, most
approaches focus on candidate regions like edge, contract etc
in segmentation process. It might segment non plate regions
rather than plate regions. So VQ based approach is able to
solve this problem. As per this paper a license plate is
assumed be a rectangular box and each rectangle regions are
considered as stripes. By using four-step codebook algorithm,
a license plate is segmented. The system was developed
under C++ language with Intel Pentium board running at 200
MHz having Windows NT operating system. The overall
processing time including acquisition, location and OCR is
about 200ms.
In [43], license plate character extraction for video
is discussed. As per Cui, Yuntao, localization of license plate
means finding text in the images. The authors assumed license
plate with light background and characters with dark back
ground. To do localization, spatial variance method is used for
finding text regions and non text regions as high variance
means text region and low variance means non text region.
In [44] DT-CNN and classifier combining based
VIPUR system (Vehicle Identification on PUblic Roads) is
described. The system does automatic recognition of license
plate for toll collection. The input is 8 bit gray scale image. As
per Ter Brugge, M.H. et al. a license plate is rectangle area
containing number of dark characters. To determine license
plate region, greyness and texture features are applied to each
pixel. To extract greyness feature DT-CNN is used which
mentions that a 1 value indicates gray value and -1 indicates
color of pixel is outside of the gray value. By using gray value
a frequency table is prepared. Based on this table a value
range is defined. The authors mentioned that gray value of
most of the license plate characters ranges from 0.1 to 0.57.
For texture feature a 3 X 3 Sobel operator is applied to each
pixel to construct histogram. Most of the license plate pixels
have absolute Sobel value more than 0.73. Then by using a
two layer DT-CNN a 3 X 3 matrix is prepared, which is
remitted to the next step for further processing. Next step is
dimension based objection selection for locating license plate
regions. In this method non plate regions are removed by
considering weak size constraints minimum height, maximum
height, minimum width and maximum width. This is done by
four DT-CNNs. Finally one union and two subtraction
operators are used for getting segmented license plate region.
In [45], pre-processing, segmentation and
verification are used for number plate localization of car. In
the pre-processing the global Thresholding is used to map
color intensity into gray scale. In plate’s vertical boundary
detection stage, Robert’s edge operator is used to emphasize
vertical edges. The authors assumed two points: plates are
oriented horizontally and there is a significant intensity
difference between plate background and character
foreground. A horizontally oriented rank-filter of size M × N
pixels is moved alongside with the image. Then by using
vertical projection is used to detect plate’s vertical boundaries.
Sometimes the plate is skewed as shown in fig. 2(a). To
eliminate skew position in a number plate a Randon
Transform (RT) function is used in conjunction with Dirac’s
delta function. Horizontal boundaries are detected by using
series of morphological erosions with horizontally oriented
structured elements are applied. Finally two step verification
procedure is used. In the first step, the global binarization is
done via Otsu’s algorithm. The authors assumed dark
characters with light background for this purpose. In the
second step, the authors compared binarization threshold with
the plate intensity median. As per Shapiro et al. probability of
detecting number plate is higher when the intensity median of
the plate zone is greater than the threshold. After passing all
the test number plate is approved successfully. If any of the
tests fails then current plate region is rejected and the system
goes back to the segmentation stage to search for the another
plate region. After maximum number of iteration if system is
not able to find plate region then an algorithm is stopped and
error message is issued. The average processing time for
single image was 0.4s on 3 GHz Pentium processor with a
varying distance of 3 to 10m from camera to vehicle.

25
A dynamic programming based plate
segmentation algorithm is proposed in [46]. The authors
discussed problem of plate segmentation and provided
different approach to solve it. Similar approach is proposed in
[47]. The authors used multiple thresholds method to extract
candidate regions. The authors used segmented blobs to
provide geometric constraints for numeric characters of a
number plate. So it is not required use any image features
likes edges, colors, or lines. The experiments were carried out
on on a desktop computer equipped with an Intel Core-Duo
3.0GHz CPU. The authors used Adaboost [48] algorithm with
OpenCV to train. It took 0.5s to detect and locate number
plate. The system is mainly used to detect Korean number
plate characters.
In [49], a fuzzy discipline based approach is
proposed for number plate segmentation. In license plate
locating module, the authors considered number plates having
colors white, black, red and green. The edge detector
algorithm is sensitive to only black-white, red-white and
green-white edges. The transformation from RGB to HIS is
used based on the formula given in this paper. After that
different edge maps are formed, a two stage fuzzy aggregator
is used to integrate these maps. After that plate candidates are
determined based on the integrated map. By using color edge
detector, fuzzy maps and fuzzy aggregation the candidate
region is located. Before remitting number plate for further
processing binarization, connected component and noise
removal is applied to it. The algorithm was executed on
Pentium - IV 1.6 GHz PC. It took around 0.4s to locate all the
possible license plate regions.
A novel future fusion approach is proposed in
[50]. The authors converted color image into gray scale by
using multiple thresholds and Otsu’s threshold. The authors
used different approaches such as deterministic approach:
pixel voting, probabilistic approach: global binarization,
probabilistic approach: local binarization and combination
among these approaches to locate the license plate.
In [51], to detect license plate candidate region
verification process is proposed. It consists of verification of
lower part, upper part and vertical border.
In [52], a cognitive and video-based approach is
proposed. As per Thome et al. license plate detection
approach falls in two groups: appearance based and gradient
methods. The appearance based approach depends on color or
texture feature to separate license plate from background.
Gradient based approach is used for high contrast license
plate. The authors used gradient-based strategy for license
plate detection. Each frame is thus processed to localize areas
with a high vertical gradient density. A vertical Sobel mark
filter is applied after contour enhancement. The pixels are
identified as text or non text area by using final labelling.
Then connected component analysis (CCA) is applied on
binary map to extract set of N candidates for license plate.
The experiment was carried out on Pentium processor with
2.66GHz with 512MB RAM. The software was developed
using Microsoft Visual studio 2005 without code
optimization. Processing time for license plate detection was
15ms.
An inductive learning RULE-3 based system is
proposed in [20]. RULE-3 is a simple algorithm containing
several steps for extracting objects having certain attributes.
Recognition of number plate contains four steps such as
finding edges which are contained in letter,
Recognizing the letter using an extracted set of rules,
Applying previous steps to all characters contained in the
number plate being processed and Recognize number plate by
bringing all characters used together.
In [23], fuzzy-based algorithm is applied. To extract
license plate region a four step method is implemented. In the
first step noise is eliminated from the input image. Edge
detection is used in second step of find rectangle area of
candidate region. In the third step, based on size, histogram
and other information invalid rectangle areas are discarded. In
the last step geometric rectification is used to obtain license
plate candidate region. As these steps need some addition
processing, authors used fuzzy-based algorithm containing
several steps to extract license plate with more accuracy. The
system was developed on TI DM642 600 MHz/32 MB RAM
with C language under CCStudio V3.1 environment. The
overall average processing time was ~418.81ms.
Number plate can be detected based on color, edge
and other features. An edge-based color-aided method is
discussed in [53]. For image enhancement intensity variance
and edge density methods are proposed. After applying
intensity variance and edge density methods, license plate
detection process is executed. The plate detection involves
several steps. Vertical edge density estimation is used to avoid
missing plate edges due to bad lighting. In the next step a
Gaussian mixture is used to emphasize the constancy of
intensity values within plate region, along horizontal
direction. Then 80% threshold value is used to detect
candidate region. Finally morphological processing and color
analysis of candidate regions are used to segment license
plate. The system was developed using MATLAB on Pentium
3.0 GHz. The overall average processing time was ~1.36s.
A probabilistic neural network (PNN) based
approach is presented in [11]. The PNN algorithm works on
gray scale image. Bottom-Hat filtering is used to enhance the
potential plate regions. To separate the object of interest from
background a thresholding is employed for binarization of the
gray level image. Because of varying lighting conditions,
brightness levels may vary and some adaptation is necessary.
To perform it Otsu's Thresholding technique is used as it is
adaptive in nature. Each segment of the binary image is
labelled according to color of each segment to enable
classification. The plate extraction is done calculating the
Column Sum Vector (CSV) and its local minima. The
algorithm was executed on Intel® Core™2 Duo Processor
CPU P8400 (2.26GHz, 2267 MHz). The plate recognition
processing time was 0.1s.
In [24], a robust and reliable SIFT based method is
used to describe local feature of a number plate. As per
Morteza Zahedi et al., the set of features extracted from the
training images must be robust to changes in image scale,
noise and illumination in order to perform reliable
recognition. The more details regarding license plate
recognition are not mentioned in this paper.
A MATLAB based system is proposed in [54]. The
authors used pre-processing step to convert color image into
binary based on the equation presented in this paper. Then by
using adaptive median filter, vehicle images gray stretch and
vehicle images gradient sharpening a license plate is
extracted.
In [28], a trichromatic imaging and color-discrete
characteristic approach is used to locate license plate. As per
Xing Yang et al. number plates from 105 countries are
composed of 11 combinations such as : (1) cyan–black; (2)

26
cyan–white; (3) black–red; (4) black–white; (5) blue–white;
(6) white–red; (7) white–green; (8) yellow–black; (9) yellow–
blue; (10) yellow–green; and (11) yellow–red. Based on this
information, authors divided six groups of RGB components.
Then authors derived several trichromatic functions and a
binarized image is obtained. In the next step, de-noising and
searching are used to locate license plate finally. The system
was Intel Pentium 4, 2.4 GHz, and 1-GB memory. The overall
average processing time was 57ms.
Image segmentation techniques such as color
texture based [55], coarse-to-fine strategy[56], wrapper based
approach[57], content based image retrieval [58], dynamic
region merging [59], Dual Multiscale Morphological
Reconstructions and Retrieval Applications [37], background
recognition and perceptual organization [60], niching particle
swarm optimization [61], constraint satisfaction neural
network [62], two stage self organizing network [63],
adaptive local thresholds [64], vectorial scale-based fuzzy-
connected image segmentation [65], mixed deterministic and
Monte-Carlo [66], evaluation matrix based image
segmentation [67], neutrosophic set and wavelet
transformation [36], non linear distance matrix [68], shape-
prior based image segmentation with intensity-based image
registration [69] and least squares support vector machine
(LS-SVM) [70] can be useful for object detection. In [71],
survey of different image segmentation techniques is
discussed. The authors focused on different unsupervised
methods. Some of these methods such as [56], [36] and [70]
can be applicable to number plate detection. The method
discussed in [56] can be useful to detect multiple objects from
the image. It is more suitable for segmentation of multiple
vehicle number plates from the traffic image. The work
Table 1. Number plate detection rate and image size
Ref Image size Success
Rate (in %)
Ref Image Size Success
Rate (in %)
[5] 1024 X 768 96,5 [43] 40 X 280 Not reported
[33] 640 X 480 Not reported [44] 640 X 480 Not reported
[17] 720 X 576
1920 X
1280
90,1 [45] Not reported 81,20
[34] Not
reported
87 [49] 640 X 480
768 X 512
97,9
[15] 640 X 480 97,3 [47] 692 X 512 97,14(Four Characters)
[13] 236 X 48 Not reported [50] 480 X 640 61,36(Pixel voting)
90,65 (Global Thresholding)
78,26 (Local Thresholding)
93,78 (Combination of global and local
binarization)
[19] 640 X 480 97,16 [51] 1300 X 1030 92,31
[39] 360 X 288
to
1024 X 768
94 [52] 640 X 480 98,3
[6] 220x50 98,82 [21] 324 X 243 97,6
[14] 648 × 486 96,4 [23] 720 X 576
768 X 576
Not reported
[8] 640 X 480 89 [53] 640 X 480 ~75,17
[41] 640 X 480 Not reported [11] 384 X 288 91
[42] 768 X 256 87,6 [28] 600 X 330
768 X 576
94,7
[38] 640 X 480 Not reported
Bold indicates overall success rate is mentioned but number plate detection rate is not mentioned

27
presented in [46], [72] lack description and clarity regarding
technical details of number plate segmentation algorithm, A
Comparative study of image segmentation techniques and
object matching using segmentation is well explained in
[73].
2.8 Discussion
In most of the literatures, the number plate segmentation
algorithms work in restricted conditions like illumination,
number plate shape (generally rectangle), size, distance from
camera and vehicle and color. It is to be noted that only few
algorithms work for real-time video image of a number plate
[33], [30],[39], [51], [61], otherwise static image of number
plate is remitted to ANPR for further processing. In Table 1,
different plate segmentation detection success rate against
plate resolution of different ANPR is depicted. The systems in
which image size and success rate of number plate detections
are not mentioned, are not included in Table 1. It is observed
that plate segmentation processing time is ranged from 15ms
to 1360ms. The lower processing time of 15ms was reported
in [52] while higher processing time of 1360 was reported in
[53]. It is evident that number plate detection rate affects
character segmentation and character recognition which in
turn affects overall recognition rate. Based on the study of
several literature presented in this section, it can be concluded
that Image binarization, Sliding concentric window (SCW),
Sobel operator, Canny-edge operator, Connected component
analysis(CCA), Hough Transform (HT), Fuzzy discipline
based approach, Probabilistic neural network (PNN) and
trichromatic imaging with color-discrete characteristic
approach can provide promising result for number plate
segmentation.
In order to proceed with character recognition,
further image processing in the form of character
segmentation is required, which is discussed in the next
section.
3. CHARACTER SEGMENTATION
After locating number plate, characters are examined for the
further process. As with the plate segmentation there are
various methods available for conducting character
segmentation. As many methods fall in more than one
category it is not possible to do category wise discussion. In
this section common related work in this area followed by
discussion is discussed. Some methods such as image
binarization and CCA are already discussed in Section 2 can
also be applied to character segmentation.
3.1 Related work in character
segmentation
In [5], the candidate region is cropped in 78 X 228
pixels by using bicubic interpolation and then subjected to
SCW for segmentation. The authors used threshold value of
0.7 for optimization of the results. After the character
segmentation process, each character is resized to pixel size of
9 X 12.
Prathamesh Kulkarni et al. [34] conclude that blob
coloring and peak-to-valley methods are not suitable for
Indian number plate. The authors proposed image scissoring
algorithm in which a number plate is vertically scanned and
scissored at the row where there is no white pixel and this
information is stored in the matrix. In case of more than one
matrix, a false matrix is discarded based on the formula given
in this paper. Same process is repeated for horizontal direction
by taking width as a threshold.
CCA is very useful technique for processing binary
image. In [19] horizontal and vertical correction and image
enhancement are performed as pre-processing steps for
character segmentation. CCA is used in horizontal and
vertical correction. After performing these steps plate is
transformed to black characters / white background and then
resized to 100 X 200. Then all the characters are segmented to
the unique size of 32 X 32. In [44] image binarization and
connected component labelling methods are used
In [16], three matrices are used to storing plate
location and binarization, number of columns in BW and
number of row in BW respectively. Then after precise
location of top and bottom boundaries are detected, which are
followed by vertical projection and Thresholding to segment
the characters.
H.Erdinc Kocer [6] used contrast extension,
median filtering and blob coloring methods for character
segmentation. Contrast extension is used to make image
sharp. As per H.Erdinc Kocer the histogram equalization is a
popular technique to improve the appearance of a poor
contrasted image. In median filtering unwanted noisy regions
are removed. Blob coloring method is applied to binary image
to detect closed and contact less regions. In this method, an L
shaped template is used to scan image from left to right and
top to bottom. This scanning process is used to determine the
independent regions by obtaining the connections into four
directions from zero valued background. The four directional
blob coloring algorithm is applied to the binary coding license
plate image for extracting the characters. At the end of this
process the numbers are segmented in the size of 28 X 35 and
letters are segmented in the size of 30 X 40. Another
algorithm based on blob detection is proposed in [14]. The
character segmentation process consists of character height
estimation, character width estimation and blob extraction.
Character height estimation contains three parts: color reverse,
vertical edge detection and horizontal projection histogram.
Color reverse is used to make color of license plate characters
as black by using statistical analysis of edges. Vertical edge
detection is used to detect finalized number plate. Sobel mask
and image binarization algorithms are used to perform it.
Horizontal projection histogram is used to find top and bottom
boundary of a character. The distance between upper and
lower boundaries is considered as height of a character.
Character width estimation contains: image binarization and
vertical projection histogram. Image binarization is used to
make color as black and white. Vertical projection is used to
find gaps between characters. The process is similar as
horizontal projection. Blob extraction is a two- step procedure
including Blob detection and blob checking algorithms. The
blob detection algorithm is an extension of CCA. Blob
checking is used to remove non blob characters from the
segmented characters.
In [45], character clipper is used to separate
character in rectangle box. Then by using feature extractor,
classifier, post processor and training phases each character is
segmented.
In [25], an improved projection method (IPM) is
proposed. The authors mentioned a three step procedure for
character segmentation. In first step horizontal, vertical and
compound tilts are corrected. Then in second steps auxiliary
lines are drawn in between first and last character to detected
connected boundaries. In the final step the characters are

28
segmented after removing noise. The experiment was carried
out using MATLAB 6.5 using VC++ 6.0.
As per Thome, Nicolas et al. [52] connected component
labelling is accurate but might result in failure because of
single mis-labelled pixel error. The authors also concluded
that histogram projection is more robust but it is less accurate.
They used connected component labelling method to
preliminary set of contours.
Vertical and horizontal histogram methods are used in [11].
To determine the boundaries of characters, column sum
vectors are used. Based on the algorithm two adjacent
characters are separated in two.
3.2 Discussion
Character segmentation is very important in order to perform
character recognition with good amount of accuracy.
Sometimes character recognition is not possible due to error
in character segmentation. In some literature of ANPR,
character segmentation is not discussed with details. Some
methods such as image binarization, CCA, vertical and
horizontal projection can produce better results of character
segmentation.
4. CHARACTER RECOGNITION
As discussed in Section 2, character recognition helps in
identifying and converting image text into editable text. As
most of the number plate recognition algorithms use single
method for character recognition. In this section, each method
is explained.
4.1 Artificial Neural Network (ANN)
Artificial Neural Network (ANN) sometimes known as neural
network is a mathematical term, which contains
interconnected artificial neurons. Several algorithms such as
[5], [6], [7], [8], [18], [52], [11], [10] are based on ANN. In
[5] two layer probabilistic neural network with the topology
of 180-180-36. The character recognition process was
performed in 128ms. In [6] multi layered perceptron (MLP)
ANN model is used for classification of characters. In
contains input layer for decision making, hidden layer to
compute more complicated associations and output layer for
resulting decision. Feed forward back-propagation (BP)
algorithm was used to train ANN. BP neural network based
systems are proposed in [8], [18], [10] with the processing
time of 0.06s, in [27]. In [46], HNN is applied to reduce
ambiguity between the similar characters e.g. 8 and B, 2 and Z
etc. The authors claim to have more than 99% recognition
rate.
4.2 Template matching
Template matching is useful for recognition of fixed sized
characters. It can be also used for detection of objects
generally in face detection and medical image processing. It
is further divided in two parts: feature based matching and
template based matching. Feature based approach is useful
when template image has strong features otherwise template
based approach can be useful. In [34] statistical feature
extraction method is applied for achieving 85% of character
recognition rate. In [15], several features and extracted and
salient is computed based on training characters. A linear
normalization algorithm is used to adjust all characters with
uniform size. The recognition rate of 95.7% is achieved
among 1176 images. An SVM based approach is used for
feature extraction of Chinese, Kana and English, Numeric
characters. The authors achieved success rate of 99.5%,
98.6%, and 97.8% for numerals, Kana, and address
recognition respectively. A template based approach is
proposed in [16]. The authors used low-resolution template
matching method to work with lower resolution image such as
4 X 8. The authors used similarity function to measure
similarity between patterns.
4.3 Other methods
In some algorithms character recognition is done by the
available Optical Character Recognition (OCR) tool. There
are numerous software available for OCR processing. One of
the open source OCR tools with multilingual supported is
Tesseract [74], [75], which is maintained by Google and
available at [76]. It is used in [14] for character recognition.
The author modified it to achieve 98.7% of character
recognition rate. In [43] authors model extraction of
characters as Markov Random Fields (MRF) where the
randomness is used to model the uncertainty in the assignment
of the pixels. Character extraction is done as the optimization
problem based on prior knowledge to maximize a posteriori
probability. Then a greedy mutation operator is used to reduce
computation cost. The method proposed in [49] consists of
three steps: character categorization, topological sorting and
self organizing (SO) recognition. Character categorization is
used to classify character as alphabet or number. In second
step topological features of input character are computed and
compared with already stored character templates.
Compatible templates will form a test set, in which the
character template that best matches the input character is
determined. The template test is performed by a SO character
recognition method Self-organized neural network is based on
Kohonen’s self-organized feature maps to handle noisy,
broken, or incomplete characters. To differentiate the similar
characters from character pairs such as (8, B) and (O, D) the
authors predefined an ambiguity set containing the characters
0, 8, B and D. For each character in the set, the non-
ambiguous parts of the character are specified as it is shown
in fig. 3. After the unknown character is classified as one of
the characters in the ambiguity set, a minor comparison
between the unknown character and the classed character is
performed. Then in the comparison process only non
ambiguous parts of characters are focused. The authors
achieved recognition rate of 95.6% for the upright license
plate images. In [77] a survey on automatic character
recognition method is discussed. As per Anju K Sadasivan
and T.Senthilkumar [77] the main challenge in character
recognition is to handle unknown text layout, different font
sizes, different illumination conditions, reflections, shadowing
and aliasing.
4.4 Discussion
Fig.3. Distinguishing parts of ambiguous characters in [49].
.

29
As character segmentation is the pre-processing steps of
character recognition, the recognizer system should be able to
handle ambiguous, noisy or distorted characters received from
character segmentation phase. Good results are reported with
ANN and self organizing (SO) recognition. The details are
summarized in Table 2. As OCR is widely used and popular
method recently, ANPR developers are focusing on improving
accuracy of OCR rather than to redesign the entire ANPR
from the scratch. As it is discussed in previous section, some
developers are using open source OCR such as Tesseract and
modifying it for better accuracy.
5. CONCLUSION
5.1 Future work
ANPR can be further exploited for vehicle owner
identification, vehicle model identification traffic control,
vehicle speed control and vehicle location tracking. It can be
further extended as multilingual ANPR to identify the
language of characters automatically based on the training
data It can provide various benefits like traffic safety
enforcement, security- in case of suspicious activity by
vehicle, easy to use, immediate information availability- as
compare to searching vehicle owner registration details
manually and cost effective for any country For low
resolution images some improvement algorithms like super
resolution [30], [31] of images should be focused. Most of the
ANPR focus on processing one vehicle number plate but in
real-time there can be more than one vehicle number plates
while the images are being captured. In [5] multiple vehicle
number plate images are considered for ANPR while in most
of other systems offline images of vehicle, taken from online
database such as [78] are given as input to ANPR so the exact
results may deviate from the results shown in Table 1 and
Table 2. To segment multiple vehicle number plates a coarse-
to-fine strategy [56] could be helpful.
Table 2. Character recognition rate with method and type of category
Ref Method Success
Rate (in
%)
Type of Category
[5] Two layer PNN 89,1 Letters
[34] Statistical feature extraction. 85 Not reported
[15] Feature salient 95,7 Not reported
[19] SVM Integration with feature extraction 93,7 English characters,
Chinese, Numeral,
Kana
[16] Template matching Not
reported
Letters, digits
[6] multi layered perceptron (MLP) ANN 98,17 Letters, digits
[7] multi layered perceptron (MLP) ANN Not
reported
Letters, digits
[14] Open source OCR Tesseract 98,7 Letters, digits
[8] BP neural network Not
reported
Korean Letters, digits
[18] BP neural network Not
reported
Chinese letters, English letters
and digits
[43] MRF Not
reported
Letters, digits
[49] character categorization, topological sorting, and
self-organizing (SO) recognition
95,6 Letters, digits
[52] Hierarchical Neural Network(HNN) 95,2 Letters, digits
[11] PNN 96,5 Letters, digits
[10] BP neural network 93,5 Letters, digits

30
5.2 Summary
It is quite clear that A]NPR is difficult system because of
different number of phases and presently it is not possible to
achieve 100% overall accuracy as each phase is dependent on
previous phase. Certain factors like different illumination
conditions, vehicle shadow and non-uniform size of license
plate characters, different font and background color affect the
performance of ANPR. Some systems work in these restricted
conditions only and might not produce good amount of
accuracy in adverse conditions. Some of the systems are
developed and used for specific country, which is summarized
in table 3. The systems in which there is no mention of
country are not included in table 3. As per table 3, it is
evident that very few of the ANPR [34], [7] are developed for
India. So there is a wide scope to develop such system for the
country like India.
This paper provides comprehensive study of recent
development and future trends in ANPR, which can be helpful
to the researchers who are involved in such developments.
6. ACKNOWLEDGMENTS
The authors thank Mr. Dharmendra Patel for providing
necessary guidance in conducting this research. The authors
also thank Charotar University of Science and Technology
(CHARUSAT) for providing necessary resources for
conducting this survey.
7. REFERENCES
[1] You-Shyang Chen and Ching-Hsue Cheng, "A Delphi-
based rough sets fusion model for extracting payment
rules of vehicle license tax in the government sector,"
Expert Systems with Applications, vol. 37, no. 3, pp. 2161-
2174, 2010.
[2] Anton Satria Prabuwono and Ariff Idris, "A Study of Car
Park Control System Using Optical Character Recognition
," in International Conference on Computer and Electrical
Engineering, 2008, pp. 866-870.
[3] A Albiol, L Sanchis, and J.M Mossi, "Detection of Parked
Vehicles Using Spatiotemporal Maps," IEEE Transactions
on Intelligent Transportation Systems, vol. 12, no. 4, pp.
1277-1291, 2011.
[4] Christos Nikolaos E. Anagnostopoulos, Ioannis E.
Anagnostopoulos, Ioannis D. Psoroulas, Vassili Loumos,
and Eleftherios Kayafas, License Plate Recognition From
Still Images and Video Sequences: A Survey, vol. 9, no. 3,
pp. 377-391, 2008.
[5] Christos Nikolaos E. Anagnostopoulos, Ioannis E.
Anagnostopoulos, Vassili Loumos, and Eleftherios
Kayafas, "A License Plate-Recognition Algorithm for
Intelligent Transportation System Applications," pp. 377-
392, 2006.
[6] H. Erdinc Kocer and K. Kursat Cevik, "Artificial neural
netwokrs based vehicle license plate recognition,"
Procedia Computer Science, vol. 3, pp. 1033-1037, 2011.
[7] A Roy and D.P Ghoshal, "Number Plate Recognition for
use in different countries using an improved
segmenation," in 2nd National Conference on Emerging
Trends and Applications in Computer
Science(NCETACS), 2011, pp. 1-5.
[8] Kaushik Deb, Ibrahim Kahn, Anik Saha, and Kang-Hyun
Jo, "An Efficeint Method of Vehicle License Plate
Recognition Based on Sliding Concentric Windows and
Artificial Neural Network," Procedia Technology, vol. 4,
pp. 812-819, 2012.
[9] Lucjan Janowski et al., "Quality assessment for a visual and
automatic license plate recognition," Multimedia Tools
and Applications Springer US, pp. 1-18, 2012.
[10] Yifan Zhu, Han Huang, Zhenyu Xu, Yiyu He, and Shiqiu
Liu, "Chinese-style Plate Recognition Based on Artificaial
Neural Network and Statistics," Procedia Engineering,
vol. 15, pp. 3556-3561, 2011.
[11] Fikriye Öztürk and Figen Özen, "A New License Plate
Recognition System Based on Probabilistic Neural
Networks," Procedia Technology, vol. 1, pp. 124-128,
2012.
[12] Jian Liang, D Dementhon, and D Doermann, "Geometric
Rectification of Camera- Captured Document Images,"
IEEE Transactions on Pattern Analysis and Machine
Intelligence, vol. 9, no. 3, pp. 591-605, 2008.
[13] Xin Fan and Guoliang Fan, "Graphical Models for Joint
Segmentation and Recognition of License Plate
Characters," IEEE Signal Processing Letters, vol. 16, no.
1, pp. 10-13, 2009.
[14] Lihong Zheng, Xiangjian He, Bijan Samali, and Laurence
T. Yang, "An algorithm for accuracy enhancement of
Table 3. Different ANPR systems with country supported
Ref Country(In which ANPR is
applied)
[5] European
[17] USA, China, Singapore, Australia, South
Africa
[34] India
[15],[35],[18],[23],
[22],[40]
China
[7] Nigeria, Cyprus, Denmark, Germany,
Estonia, Finland, France, India, Norway,
Slovakia, Portugal, U.S.A, Bulgaria,
Czech Republic
[44] Dutch
[45] Israel, Bulgaria
[47],[26],[51] Korea
[52] Multi-country
[20] Turkey
[21] Australia
[53],[24] Iran
[27] USA
[28] China and 104 Countries

31
license recognition," Journal of Computer and System
Sciences, , 2012.
[15] Zhen-Xue Chen, Cheng-Yun Liu, Fa-Liang Chang, and
Guo-You Wang, "Automatic License-Plate Location and
Recognition Based on Feature Saliance," IEEE
Transactions on Vehicular Technology, vol. 58, no. 7, pp.
3781-3785, 2009.
[16] Ch.Jaya Lakshmi, Dr.A.Jhansi Rani, Dr.K.Sri
Ramakrishna, and M. KantiKiran, "A Novel Approach for
Indian License Recognition System," International
Journal of Advanced Engineering Sciences and
Technologies, vol. 6, no. 1, pp. 10-14, 2011.
[17] Jianbin Jiao, Qixiang Ye, and Qingming Huang, "A
configurabe method for multi-style license plate
recognition," Pattern Recognition, vol. 42, no. 3, pp. 358-
369, 2009.
[18] Zhigang Zhang and Cong Wang, "The Reseach of Vehicle
Plate Recogniton Technical Based on BP Neural
Network," AASRI Procedia, vol. 1, pp. 74-81, 2012.
[19] Ying Wen et al., "An Algorithm for License Plate
recognition Applied to Intelligent Transportation System,"
IEEE Transactions of Intelligent Transportation Systems,
pp. 1-16, 2011.
[20] Mehmet Sabih Aksoy and Ahmet Kürsat Türker Gültekin
Çagıl, "Number-plate recognition using inductive
learning," Robotics and Autonomous Systems, vol. 33, no.
2-3, pp. 149-153, 2000.
[21] Wenjing Jia, Huaifeng Zhang, and Xiangjian He, "Region-
based license plate detection," Journal of Network and
Computer Applications, vol. 30, no. 4, pp. 1324-1333,
November 2007.
[22] Yang Yang, Xuhui Gao, and Guowei Yang, "Study the
Method of Vehicle License Locating Based on Color
Segmentation," Procedia Engineering , vol. 15, pp. 1324-
1329, 2011.
[23] Feng Wang et al., "Fuzzy-based algorithm for color
recognition of license plates," Pattern Recognition Letters,
vol. 29, no. 7, pp. 1007-1020, May 2008.
[24] Morteza Zahedi and Seyed Mahdi Salehi, "License plate
recognition system based on SIFT features," Procedia
Computer Science, vol. 3, pp. 998-1002, 2011.
[25] Mei-Sen Pan, Jun-Biao Yan, and Zheng-Hong Xiao,
"Vehicle license plate character segmentation ,"
Intenational Journal of Automation and Computing, pp.
425-432, 2008.
[26] Kaushik Deb, Andrey Vavilin, Jung-Won Kim, and
Kang-Hyun Jo, "Vehicle license plate tilt correction
based on the straight lne fitting method and minimizing
variance of coordinates of projection point,"
International Journal of Control, Automation and
Systems., pp. 975-984, 2010.
[27] Francisco Moraes Oliveira-Neto, Lee D. Han, and
Myong K Jeong, "Online license plate matching
procedures using license-plate recognition machine and
new weighted edit distance," Transportation Research
Part C: Emerging Technologies, vol. 21, no. 1, pp. 306-
320, April 2012.
[28] Xing Yang, Xiao-Li Hao, and Gang Zhao, "License plate
location based on trichromatic imaging and color-
discrete characteristic," Optik- International Journal for
Light and Electron Optics, vol. 123, no. 16, pp. 1486-
1491, August 2012.
[29] Cynthia Lum, Julie Hibdon, Breanne Cave, Christopher
S. Koper, and Linda Merola, "License plate
reader(LRP) police patrols in crime hot spots: an
experimental evaluation in two adjacent jurisdictionss,"
Journal of Experimel Criminology, Springer
Netherlands, , pp. 321-345, 2011.
[30] K.V. Suresh, G. Mahesh Kumar, and A.N. Rajagopalan,
"Superresolution of license plates in real traffic videos,"
IEEE Trans. Intell. Transp. Syst, vol. 8, no. 2, pp. 321-
331, 2007.
[31] Yushuang Tian, Kim-Hui Yap, and Yu He, "Vehicle
license plate super-resolution using soft learning prior,"
Multimedia Tools and Applications, Springer US, pp.
519-535, 2012.
[32] D.H. Ballard, "Generalizing the Hough Transform to
Detect Arbitary Shapes," Pattern Recognition, vol. 13,
no. 2, pp. 111-122, 1981.
[33] Shen-Zheng Wang and Hsi-Jian Lee, "A cascade
framework for real-time statistical plate recognition
system," IEEE Trans. Inf. Forensics security, vol. 2, no.
2, pp. 267-282, 2007.
[34] Prathamesh Kulkarni, Ashish Khatri, Prateek Banga, and
Kushal Shah, "Automatic Number Plate Recognition
(ANPR)," in RADIOELEKTRONIKA. 19th International
Conference, 2009.
[35] Hui Wu and Bing Li, "License Plate Recognition
System," in International Conference on Multimedia
Technology (ICMT), 2011, pp. 5425-5427.
[36] Abdulkar Sengur and Yanhui Guo, "Color texture image
segmentation based on neutrosophic set and wavelet
transformation ," Computer Vision and Image
Understanding, vol. 115, no. 8, pp. 1134-1144, August
2011.
[37] Jiann-Jone Chen, Chun-Rong Su, W.E.L Grimson, Jun-
Lin Liu, and De-Hui Shiue, "Object Segmentation of
Database Images by Dual Multiscale Morphological
Reconstructions and Retrieval Applications," IEEE
Transactions on Image Processing, vol. 21, no. 2, pp.
828-843, 2012.
[38] Mahmood Ashoori Lalimi, Sedigheh Ghofrani, and Des
McLernon, "A vehicle license plate detection method
using region and edge based methods," Computers &
Electrical Engineering, November 2012.
[39] M. S. Sarfraz et al., "Real-Time automatic license plate
recognition for CCTV forensic applications," Journal of
Real-Time Image Processing- Springer
Berlin/Heidelberg, 2011.
[40] Rongbao Chen and Yunfei Luo, "An Improved License
Plate Location Method Based On Edge Detection,"
Physics Procedia, vol. 24, pp. 1350-1356, 2012.
[41] T Naito, T Tsukada, K Kozuka, and S yamamoto,
"Robust license-plate recognition method for passing
vehicles under outside environment," IEEE
Transactions on Vehicular Technology, vol. 49, no. 6,
pp. 2309-2319, 2000.

32
[42] R Zunino and S Rovetta, "Vector quantization for
license-plate location and image coding," IEEE
Transactions on Industrial Electornics, vol. 47, no. 1,
pp. 159-167, 2000.
[43] Yuntao Cui and Qian Huang, "Extracting character of
license pltes from video sApplicationsequences,"
Machine Vision and Applications, Springer Verlag, p.
308, 1998.
[44] M.H Ter Brugge, J.A.g jhuis, L Spaanenburg, and J.H
Stevens, "CNN- Applications in Toll Driving," Journal
of VLSI signal processing systems for signal, image and
video tehnology, pp. 465-477, 1999.
[45] Vladimir Shapiro and Georgi Gluhchev Dimo Dimov,
"Towards a Multinational Car License Plate
Recognition system," Machine Vision and
Appplcations, Springer-Verlag, pp. 173-183, 2006.
[46] E.N Vesnin and V.A Tsarev, "Segmentation of images of
license plates," Pattern Recogniton and Image Analysis,
pp. 108-110, 2006.
[47] A Kang, D. J;, "Dynamic programming -based method
for extraction of license numbers of speeding vehicles
on the highway ," International Journal of Automotive
Technology, pp. 205-210, 2009.
[48] P. Viola and M JOnes, "Robust real-time face detection,"
Int. J. Comput. Vis, vol. 57, no. 2, pp. 137-154, 2004.
[49] Shyang-Lih Chang, Li-Shien Chen, Yun-Chung Chung,
and Sei-Wan Chen, "Automatic license plate
recogniton," IEEE Transactions on Intelligent
Transportation Systems, vol. 5, no. 1, pp. 42-53, 2004.
[50] Rami Al-Hmouz and Subhash Challa, "License plate
location based on a probabilistic model," Machin Vision
and Applications, Springer-Verlag, pp. 319-330, 2010.
[51] J. K. Chang, Ryoo Seungteak, and Heuiseok Lim, "Real-
time vehicle tracking mechanism with license plate
recognition from reoad images," The journal of super
computing , pp. 1-12, 2011.
[52] Nicolas Thome, Antoine Vacavant, Lionel Robinault,
and Serge Miguet, "A cognitive and video-based
approach for multinational License Plate Recognition ,"
Machine Vision and Applications, Springer-Verlag, pp.
389-407, 2011.
[53] Vahid Abolghasemi and Alireza Ahmadyfard, "An edge-
based color aided method for license plate detection,"
Image and Vision Computing , vol. 27, no. 8, pp. 1134-
1142, July 2009.
[54] Xiaofeng Zhang, Fengchang Xu, and Yan Su, "Research
on the Licnese Plate Recognition based on MATLAB,"
Procedia Engineering, vol. 15, pp. 1330-1334, 2011.
[55] M. Mirmehdi and M Petrou, "Segmentation of color
textures," IEEE Transactions of Pattern Analysis and
Machine Intelligence, vol. 22, no. 2, pp. 142-159, 2000.
[56] Y. Amit, D. German, and X. Fan, "A coarse-to-fine
strategy for multiclass shape detection," IEEE Trans.
Patttern Anal. Mach. Intell, vol. 26, no. 12, pp. 1606-
1621, 2004.
[57] M.E Farmer and A.k Jain, "A wrapper based approach to
image segmentatin and classification," IEEE
Transaction on Image Processing, vol. 14, no. 12, pp.
2060-2072, 2005.
[58] Dewen Zhuang and Shoujue Wang, "Content-Based
Image Retrieval Based on Integrating Region
Segmentation and Relevance Feedback," in
International Conference on Multimedia Technology
(ICMT), 2010, pp. 1-3.
[59] Bo Peng, Lei Zhang, and D Zhang, "Automatic Image
Segmenation by Dynamic Region Merging," IEEE
Transactions on Image Processing, vol. 20, no. 12, pp.
3592-3605, 2011.
[60] Cheng, Chang; Koschan, A; Chen, Chung-Hao; Page,
D.L; Abidi, M.A, "Outdoor Scene Image Segmentation
Based on Background Recognition and Perceptual
Organization," IEEE Transactions on Image Processing,
vol. 21, no. 3, pp. 1007-1019, 2012.
[61] Shahid Mehmood, Stefano Cagnoni, Monica Mordonini,
and Shoab Ahmad Khan, "An embeded architecture for
real-time object detection in digital images based on
niching particle swarm optimization," Journal of Real-
Time Image Processing, Springer-Verlag, pp. 1-15,
2012.
[62] Fatih Kurugollu, Bülent Sankur, and A. Emre Harmanci,
"Image segmentation by relaxation using constraint
satisfaction neural network," Image and Vision
Computing, vol. 20, no. 7, pp. 483-497, May 2002.
[63] S.H. Ong, N.C. Yeo, K.H Lee, Y. V. Venkatesh, and
D.M. Cao, "Segmentation of color images using a two-
stage self-organizing network," Image and Vision
Computing, vol. 20, no. 4, pp. 279-289, 2002.
[64] Ety Navon, Ofer Miller, and Amir Averbuch, "Color
Image segmentation based on adaptive local thresholds,"
Image and Vision Computing, vol. 23, no. 1, pp. 69-85,
January 2005.
[65] Ying Zhuge, Jayaram K. Udupa, and Punam K. Saha,
"Vector scale-based fuzzy-connected image
segmentation," Computer Vision and Image
Understanding, vol. 110, no. 2, pp. 177-193, March
2006.
[66] Daniel Crevier, "Image segmentation algorithm
development using ground truth image data sets,"
Computer Vision and Image Understanding, vol. 112,
no. 2, pp. 143-159, November 2008.
[67] Mark Polak, Hong Zhang, and Minghong Pi, "An
evaluation metric for image segmentation of multiple
objects," Image and Vision Computing , vol. 27, no. 8,
pp. 1223-1227, July 2009.
[68] Antonio Carlos Sobiernaski, Eros Comunello, and Ald
von Wangenheim, "Leaning a nonlinear distance metric
for supervised region-merging image segmentation,"
Computer Vision and Image Understanding, vol. 115,
no. 2, pp. 127-139, February 2011.
[69] Siqi Chen, Daniel Cremers, and Richard J. Radke,
"Image segmentation with one shape prior - A template-
based formulation," Image and Vision Computing , vol.
30, no. 12, pp. 1032-1042, December 2012.
[70] Hong-Ying Yang, Xiang-Yang, Qin-Yan Wang, and
Xian-Jin Zhang, "LS-SVM based image segmentation
using color and texture information ," Journal of Visual

33
Communication and Image Representation , vol. 23, no.
7, pp. 1095-1112, October 2012.
[71] Hui Zhang, Jason E. Fritts, and Sally A. Goldman,
"Image Segmentation evaluation: A survey of
unsupervised methods," Computer Vision and Image
Understanding, vol. 110, no. 2, pp. 260-280, May 2008.
[72] J. Phelawan, P. Kittisut, and N. Pornsuwancharoen, "A
new technique for distance measurement of between
vehicles to vehicles by plate car using image processing,"
Procedia Engineering, vol. 32, pp. 348-353, 2012.
[73] S Sapna Varshney, N Rajpal, and R Purwar,
"Comparative Study of image segmentation techniques
and object matching using segmentation," in
International Conference on Methods adn Models in
Computer Science, 2009, pp. 1-6.
[74] Smith R, "An Overview of the Tesseract OCR Engine,"
in IEEE Ninth Intenational Conference Proceeding of
Document anay and Recognition, 2007.
[75] Chirag Patel, Atul Patel, and Dharmendra Patel, "Optical
Character Recognition by Open source OCR Tool
Tesseract : A Case Study," International Journal of
Computer Applications, Foundation of Computer
Science, New York. USA, vol. 55, no. 10, pp. 50-56,
October 2012.
[76] Google Code. [Online] (2012).
http://code.google.com/p/tesseract-ocr
[77] Anju K. Sadasivan and T. Senthilkumar, "Automatic
Character Recognition in Complex Images," Procedia
Engineering, vol. 30, pp. 218-225, 2012.
[78] Jerome Connix, License Plate Mania. [Online].
http://www.licenseplatemania.com
AUTHOR’S PROFILE
Chirag Patel received Bachelor degree in computer
applications (B.C.A) degree from Dharmsinh Desai
University Nadiad, Gujarat, India in 2002 and Masters Degree
in Computer Applications (M.C.A) from Gujarat University,
Gujarat, India in 2005. He is pursuing PhD in Computer
Science and Applications from Charotar University of Science
and Technology (CHARUSAT). He is with MCA
Department at Smt Chandaben Mohanbhai Patel Institute of
Computer Applications, Charotar University of Science and
Technology (CHARUSAT), Changa, Gujarat, India. His
research interests include Information Retrieval from
image/video, Image Processing and Service Oriented
Architecture.
Dr. Dipti Shah received Bachelor degree in Science;
B.Sc.(Maths), M.C.A. Degree from S.P. University , Gujarat,
India. She has also received Ph.D in Computer Science,
degree from S.P. University, Gujarat, India. Now she is
Professor at G.H.Patel Department of Computer Science, S.P.
University, Anand, Gujarat, India. Her Research interests
include Computer Graphics, Image Processing, Multimedia
and Medical Informatics.
Dr. Atul Patel received Bachelor in Science B.Sc
(Electronics), M.C.A. Degree from Gujarat University, India.
M.Phil. (Computer Science) Degree from Madurai Kamraj
University, India. He has received his Ph.D degree from S. P.
University.Now he is an Associate Professor and Head, Smt
Chandaben Mohanbhai Patel Institute of Computer
Applications – Changa, India. His main research areas are
wireless communication and Network Security
View publication stats
View publication stats

Automatic Number Plate Recognition System (ANPR) A Survey.pdf

More Related Content

What's hot (20)

Similar to Automatic Number Plate Recognition System (ANPR) A Survey.pdf (20)

More from Tina Gabel (20)

Recently uploaded (20)

Automatic Number Plate Recognition System (ANPR) A Survey.pdf