See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/308861417

Coin detection and recognition using neural networks

Conference Paper · March 2015

DOI: 10.1109/ICCPCT.2015.7159434

CITATIONS READS

16 713

2 authors:

Mansoor Roomi Jayanthi Rajee

Thiagarajar College of Engineering Thiagarajar College of Engineering
158 PUBLICATIONS 644 CITATIONS 6 PUBLICATIONS 31 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Jayanthi Rajee on 03 January 2023.

The user has requested enhancement of the downloaded file.

 2015 International Conference on Circuit, Power and Computing Technologies [ICCPCT]

COIN DETECTION AND RECOGNITION

USING NEURAL NETWORKS
S.Mohamed mansoor roomi R.B.Jayanthi rajee
Associate professor ,Department of ECE, Assistant professor ,Department of EEE,
Thiyagarajar college of Engineering, Velammal college of Engineering&Technology
Madurai,TN, Madurai TN,
[email protected] [email protected]

and different materials bring different changes in

Abstract:- Coin identification and recognition and the amplitude and direction of frequency. So these
is important to enhance the extended operation of
changes and the other parameters like diameter,
Vending machines, Pay phone system and coin
counting machines. Coin recognition is a difficult thickness, weight and magnetism can be used to
task in machine intelligence and computer vision differentiate between coins. The electromagnetic
problems because of its various rotations and based coin recognition systems improve the
widely changed patterns. Therefore, an efficient accuracy of recognition but still they can be fooled
algorithm is designed to be robust and invariant to by some game coins.
rotation, translation and scaling. The objective of The coin recognition systems based on
this work is to find whether the object is coin or not
image can also be divided into two categories:
if so denomination of the coin is found. The
Fourier approximation of the coin image is used to method based on image registration[3][4] and
reduce the variations on surface of coin such as method based on feature vectors with rotation
light reflection effect. Then coins can be variance[5][6][7] .The image registration is the
distinguished by feeding those features into a process of overlaying two or more images of the
multi-layered BP neural network. same scene taken at different times, from different
Keywords : object recognition; Fourier transform; viewpoints, or by different sensors. In coin
coordinate transform; BP neural network
recognition system it extracts feature points from
I.INTRODUCTION the reference image and the second image and then
uses similarity measurement to find the matched
Coin recognition systems and coin sorting pairs by which the second image is spatially
machines have become a vital part of our life. They transferred.
are used in banks, supermarkets,grocery stores,
vending machines etc. There are three types of coin II PROPOSED METHOD
recognition systems based on different methods
used by them available inthe market .(i)Mechanical The proposed method is used to identify
method based systems [1](ii),Electromagnetic coin or non coin and then find the denomination of
method based systems,(iii) Image processing based Indian coins A camera is placed inside the coin
systems [2]The mechanical method based systems drop box. The image of head side of a dropped coin
use parameters like diameter or radius, thickness, are acquired at constant illumination and fixed
But these parameters can not be used to distance with rotation invarience. A reference
differentiate between the different materials of the database is maintained withthe image of the coins
coins. It means if we provide two coins one of different variety and various denominations.
original and other fake having same diameter, First We have to find coin or non coin.The rotation
thickness, weight and magnetism but with different invariance feature is represented by the absolute
materials to mechanical method based coin value of Fourier coefficients of polar image of the
recognition system then it will treat both the coins coin on circles with different radii. Then coins can
as original coin so these systems can be fooled be distinguished by feeding those features into a
easily.The electromagnetic method based systems multi-layered BP neural network.
can differentiate between different materials
because in these systems the coins are passed
through an oscillating coil at a certain frequency

978-1-4799-7075-9/15/$31.00 ©2015 IEEE

 2015 International Conference on Circuit, Power and Computing Technologies [ICCPCT]

image to obtain the difference image (diff (x, y))

which is represented in equation (2.2.1).
diff ( x, y ) = b( x, y ) − c( x, y ) ------ 2.2.1

2.3 Edge Detection

Image Acquisition

Edges charecterize boundaries and are

therefore a problem of fundamental importance in
Segmentation image processing .The Canny edge detection
algorithm is known to many as the optimal edge
detector. The canny edge detector first smooths the
Edge Detection
image to eliminate and noise.

2.4 Hough Transform

Hough Transform This method used to find the circle in the
image. The general Hough transform can be used
on any kind of shape, although the complexity of
Polar Transform
the transformation increase with the number of
parameters needed to describing the shape. Hough
transform can be described as a transformation of a
Coin (OR) Non Coin
point in the x,y-plane to the parameter space. A
straight line passing through the points in x,y-plan
Fig 1 Proposed Method
be described by:
Y = ax + b ----- (2.4.1)

2.5 Polar Transform

2.1 Image Acquisition
The coin image is converted from
The acquisition of image to operate the Cartesian coordinates to logarithmic polar
vending machine using different denomination of coordinates. The image transform from Cartesian to
coins. A camera is fixed inside the vending logarithmic polar coordinates is introduced.
machine to capture the coin image. Illumination is Suppose the coin image can be denoted as f(x,y)
provided for image acquisition by fixing a light under the Cartesian coordinates and as f (r ,θ )
source. The light source is hinged into the vending under the logarithmic coordinates. The coordinates
machine in a proper position to have reduced transform from f(x,y) to f (r ,θ ) can be defined as:
reflection on coin.

---- (2.5.1)
2.2 Segmentation

The captured image contains the details of

the inserted coin along with unwanted details. --------(2.5.2)
Therefore, a process is needed to segment the
By applying polar transform to the hough transform
portion of the coin from the captured image. The
of image,one output will be obtained. From that we
captured image is subtracted from the background
 2015 International Conference on Circuit, Power and Computing Technologies [ICCPCT]

will decide coin or non coin.If the image is coin

Data
denomination of the coin will be found out. Segmented
Reference Base
Image
2.6 Feature Extraction

If the Image is coin means the denomination of the

coin founded out by this procedure. Feature vectors Image Acquisition
as the input for the neural network. Polar transform
Feature Extraction
is applied to original image. Then the fourier
(Fourier coefficients)
transform applied to the output of the polar Segmentation
transform .Then the feature vectors will be
obtained
Feature Extraction
Input Image
(Fourier coefficients)
Polar Transform

Fourier Transform

Feature Matching
Compute Fourier Co efefficients
(Neural network)

Taking Particular co efficients

Feature vectors Denomination of

the coin
Figure 2 Feature vector computation

2.7 Fourier Transform Fig 3.1 Schematic Flow diagram of

Denomination of the Coin
The Fourier Transform is an important image
Then obtained feature vector for given image is
processing tool which is used to decompose an
compared with the feature vector of the data base
image into its sine and cosine componentsThe
image with neural network.The neural network is
Fourier Transform is used in a wide range of
the one of the classifier to obtain good result
applications, such as image analysis, image
3.1 Neural Network
filtering, image reconstruction and image
compression. The neural network contains a large number of
simple neuron like processing elements and a large
III Proposed method to obtain Denomination of the
number of weighted connections between the
Coin
elements. The weights on the connections encode
Image acquisition & segmentation explained in the the knowledge of the network.biologically inspired,
previous section.FeatureVector computation was many of the neural network models developed do
explained in figure 2. Data base is maintained with not duplicate the operation of the human brain
coin reference images.
 2015 International Conference on Circuit, Power and Computing Technologies [ICCPCT]

3.1.1 Basic concepts The network is trained with the parameters

A neural network has a parallel distributed corresponding to various denomination of the
architecture with large number of nodes and coins, to their respective targets. After training the
connections. Each connection points from one node performance function reaches the goal for all the
to another and is associated with a weight. neural samples.
network consists of the following tasks IV Experimental setup
i)Determine the network properties : The Network The system requirements for the proposed
topology ( connectivity ) the types of connections, work are processor, camera and memory unit. The
the order of connections, and the weight range proposed work is simulated using the software on
the Intel processor of 2GHz frequency and 2GB
ii)Determine the node properties: the activation
RAM. The required images for the proposed work
range and the activation
are acquired with the help of digital camera

iii)Determine the system dynamics: the weight mounted on the cardboard box of size 40 cm x 30

initialization scheme the activation calculating cm x 30cm. The experimentation is performed in

formula and the learning rule. the closed environment, which is to match the case

3.1.2 Neuron Model of vending machines. Image acquisition in the

close environment assists to avoid the reflections
An elementary neuron with R inputs is shown in from multiple sources. A single light source is fit
figure 3.3. Each input is weighted with an into the cardboard box for illumination.
appropriate ‘w’. The sum of the weighted inputs This provides the constant illumination inside
and the bias forms the input to the transfer function the closed environment. The source used for
‘f’. lighting purpose is a CFL lamp of 11 watts power
which is flexible to rotate at different angles. The
position of the light is fixed in a position, where
less reflection of the coin is achieved. The digital
camera used for image acquisition is Sony W150.
Fig3.2Elementary neuron with ‘R’ inputs
This camera is fit at the top of the cardboard box.

3.1.3 Network Training and Classification The images acquired for processing are captured in

Multilayer feed forward network is taken as video graphics array mode (VGA).

the network architecture for the present application. 4.1 Database

After choosing the network, the number of neurons The proposed method is required to find the

in each layer has to be decided.. feature vectors. The coins for the reference images
are chosen based on the denomination and models.
The database contains 48 images. One 10 Rupees
F1 One Rupee
coin, Eleven 5 Rupees coins, eighteen 1 Rupee
F2
Two Rupee coins, seventeen 2 Rupees coins coin are used as
F3
Five Rupee the reference for experimentation.
...

F15
Ten Rupee V RESULTS & DISCUSSION

The Proposed algorithm should Check Coin or

fig 3.3Network Architecture non coin from drop box. The proposed algorithm
 2015 International Conference on Circuit, Power and Computing Technologies [ICCPCT]

uses Fourier Coefficients for feature extraction of

coin images. Here Fifteen Coefficients values are
considered for the feature vectors.

5.1 Rotation Invarient

The Coin Images taken out with different 5.2.1Original 5.2.2 Canny edge detection
rotations.The polar Transform is applied to input Image
image. Then the fourier transform is applied to that
image. Then the fourier coefficients will be
computed. In Different rotations the coin have the
more or less same fourier coefficients values. This
will be discussed below.
5.2.3 Hough Transform 5.2.4 Polar Transform

5.1.1 Original 5.1.2 Polar Transform

6
5.4 Identification of Denomination of coin
x 10
10

9 If the given input image is coin ,then the polar

transform applied to that image.Then fourier
8

6
transform applied to that transformed image.Then
5

4 the fourier coefficients extracted.Fourier

3

2
coefficients considered as the input for the neural
1
network.Then these vectors to be matched in to
0
0 50 100 150 200 250 300

those feature vectors stored in the data base .Then

5.1.3 Fourier Coefficients Values
the correct denomination of the coin identified.It
represented in fig below
5.2 Recognition Of Coin (OR) Non Coin

Segmented Output of coin is considered as the

input as the following procedure.Image
Preprocessing is done before edge detection.Then
canny edge detection is applied to image.Then the
hough transform is applied to that output image to
detect circle.Then the Statistical properties will be
computed for the hough detection output.From that
properties it will identify Coin or Non coin. Figure 5.4.1 Simulation results for denomination
of the coin(2 Rupee)
 2015 International Conference on Circuit, Power and Computing Technologies [ICCPCT]

Training of neural network for this 2 rupee is given Research inComputer Science and Software Engineering
Volume 4, Issue 1, January 2014
below
Performance is 0.275377, Goal is 0.3
2
10 [4]chen cai ming,Zhang,shi-qing,chen,yue-fen A Coin
Recognition System with Rotation invariance ,International
Conference on Machine Vision and Human-machine Interface
School of Physics and Electronic Engineering, China
Training-Blue Goal-Black

1
10

[5]Marco Reisert, Olaf Ronneberger, Hans Burkhardt, “An

Efficient Gradient based Registration Technique for Coin
0
10 Recognition”, Muscle Pattern Recognition and Image processing
group competition on CIS, 2006, doi: 10.1.1.87.4375

[6] Bennamoun J, Mamic G.J, “Object Recognition -

-1
10
0 2 4 6 8 10 12 14 16 Fundamentals and Case Studies”, Springer, 2002, ISBN
17 Epochs
185233-398-7
Figure 5.4.2 Simulation results for denomination
of the coin (2 Rupee) [7] Gonzalez, Woods, “Digital image processing”, Prentice

VICONCLUSION Hall, 2002, ISBN 0201180758.

In this work, a new coin recognition system has [8] Daisheng Luo, “Pattern Recognition and image
been proposed to identify the denomination of processing”, Horwood serries in Engineering series, ISBN-1-

Indian coins.The given input image should identify 898563-52-7

coin or non coin,After that it is identified as coin

means the feature vectors are extracted and
denomination of the coin found out with neural
network.Neural network will trained in to the
various types of coins. The recognition rate of the
proposed algorithm is around 82%. The future
work is to identification of denomination of the
coin with effectively with reduced time.

REFERENCES

[1] Sandeep Kaur, Mandeep Kaur“review on the coin

recognition system with rotation invariant”, International
Journal of Computer Science and Mobile Computing.Vol. 3,
Issue. 9, September 2014

[2] Nikita Shelgikar, Prof. L.M.R.J. Lobo “Indian Coin

Recognition with Rotation Invariance using Radial Blur
Technique” International Journal of Application or Innovation
in Engineering & Management (IJAIEM) Volume 3, Issue 1,
January 2014

[3] Suchika Malik, Parveen Bajaj, Mukhwinder Kaur“Sample

Coin Recognition System using Artificial Neural Network on
Static Image Dataset”International Journal of Advanced

View publication stats