Acknowledgements

First and foremost, I would like to thank my seminar guide, Prof. Jyoti Mankar,
for her guidance and support. I will forever remain grateful for the constant support
and guidance extended by my guide, in making this seminar report. Through our
many dis-cussions, she helped me to form and solidify ideas.
With a deep sense of gratitude, I wish to express my sincere thanks to, Prof.
Dr. S. S. Sane for his immense help in planning and executing the works in time. My
grateful thanks to the departmental sta members for their support.
I would also like to thank my wonderful colleagues and friends for listening my ideas,
asking questions and providing feedback and suggestions for improving my ideas.

Miss. Putrevu Pranati

Abstract

Monetary transactions are integral part of our day to day activities, so currency
recognition has become one of the active research area at present and it has
vast potential applications. The research has introduced a system to recognize
and classify four different currencies using computer vision.
The technology of currency recognition aims to search and extract the visible
as well as hidden marks on paper currency for efficient classification. The
features are extracted based on color, texture and shape for four different
currencies and they are classified using Artificial Neural network.

Keywords : Component, Formatting, Style, Styling, Insert

Contents

1 Introduction 1

2 Literature Survey 2
2.1 Canny Edge Detection Method . . . . . . . . . . . . . . . . . . . . . . . 2
2.2 Feedforward Back propagation Procedure . . . . . . . . . . . . . . . . . . 3

3 Structure of Typical Currency Recognition System 5

4 Proposed Approach 8
4.1 Image Aquisition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
4.2 Image Preprocessing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
4.2.1 Median Filter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
4.3 Feature Extraction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
4.3.1 LUV Color Space . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
4.4 Shape Feature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
4.4.1 Edge Detection . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
4.4.2 Gray level co-occurrence matrix . . . . . . . . . . . . . . . . . . . 12

5 Classi cation Using Arti cial Neural Network 13

5.0.1 Training Phase . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
5.0.2 Testing Phase . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14

6 Conclusion and Future Scope 17

3
References 17
Chapter 1

Introduction

Object recognition is an important and highly demanded area of pattern recog-

nition. An object can be anything in real life. It can be text in a document, a license plate
of a vehicle, an iris in a persons eyes, a sign in a sign language, a face of a person, and
so on. Similarly, paper currency recognition is as important as any other object
recognition. Some authors, in the recent years, have contributed to the subject of paper
currency recognition systems. These existing paper currency recognition methods, in
the literature, mainly involve image processing and/or neural network techniques. This
paper deals with a simple, e cient and very accurate approach in the system design. In
designing such a system, it considers di erent dimensions, Color luminance, Edge
histogram using max sobel gradient, correlations as features. A di erent method using
radial basis Function networks, is utilized for developing an intelligent system which can
recognize paper currency. This research is speci cally designed for recognizing paper
currency from United Kingdom (Pound), Japan (Yen), Europe(euro) and India(rupee). In
the proposed paper, recognition technique has been designed in such a way that it can
be used for recognizing paper currency of four di erent countries.
To overcome the problem of recognizing dirty banknotes, the pre-processing stage is
also considered. The proposed scheme is di erent from various existing methods be-cause
of its approaches in the recognition phases. Speci cally, for example, sobel masks have
been used in for considering speci c edges (images) in a paper currency. Using this method,
the features are extracted and fed to a Neural Network (NN).
1
Chapter 2

Literature Survey

2.1 Canny Edge Detection Method

The Process of Canny edge detection[2] algorithm can be broken down to 5
dif-ferent steps:
1.Apply Gaussian lter to smooth the image in order to remove the
noise 2.Find the intensity gradients of the image
3.Apply non-maximum suppression to get rid of spurious response to edge
detection 4.Apply double threshold to determine potential edges
5.Track edge by hysteresis: Finalize the detection of edges by suppressing all the
other edges that are weak and not connected to strong edges.
2
 CHAPTER 2. LITERATURE SURVEY

Fig 1 : Output of canny edge detector.

2.2 Feedforward Back propagation Procedure

FEEDFORWARD NEURAL NETWORK
A feedforward neural network[9] is an arti cial neural network wherein con-
nections between the units do not form a cycle. As such, it is di erent from recurrent
3
 CHAPTER 2. LITERATURE SURVEY

neural networks.
The feedforward neural network was the rst and simplest type of arti cial
neural network devised. In this network, the information moves in only one direction,
forward, from the input nodes, through the hidden nodes (if any) and to the output
nodes. There are no cycles or loops in the network.

FEEDFORWARD BACK PROPAGATION NEURAL NETWORK

The backward propagation of errors or backpropagation[9], is a common
method of training arti cial neural networks and used in conjunction with an optimization
method such as gradient descent. The algorithm repeats a two phase cycle, propagation
and weight update. When an input vector is presented to the network, it is propagated
for-ward through the network, layer by layer, until it reaches the output layer. The output
of the network is then compared to the desired output, using a loss function, and an error
value is calculated for each of the neurons in the output layer. The error values are then
propagated backwards, starting from the output, until each neuron has an associated
error value which roughly represents its contribution to the original output.
4
Chapter 3

Structure of Typical Currency

Recognition System

In the proposed system a high resolution scanner is used to acquire the image.
The acquired image of a paper currency is rst converted to gray scaled image. Conversion
5
CHAPTER 3. STRUCTURE OF TYPICAL CURRENCY RECOGNITION SYSTEM

to gray scale facilitates further pre-processing. The task of pre-processing is

achieved by converting colored currency images into grayscale, then black-white
images. After that, the edge of the image is ltered using Prewitt method. Then, the
image edge is detected using Canny s edge detection method.
This research is speci cally designed for recognizing paper currency from :
i) United Kingdom (Pound)
ii) Japan (Yen)
iii) United States (Dollar)
iv) India (rupee)

Pound

Yen
6
CHAPTER 3. STRUCTURE OF TYPICAL CURRENCY RECOGNITION SYSTEM

Dollar

Rupee

The goal of the paper is to achieve the best accuracy in recognizing patterns
with the lowest possible. Given the fact that paper currencies are usually recognized by
machines that have small power (such as auto-seller machines and ATMs), the cost is a
limiting factor. Therefore, it is really urgent for all paper currency recognizers to minimize
the power consumption, and, at the same time, achieve high level of accuracy.
7
Chapter 4

Proposed Approach

4.1 Image Aquisition

In this step the process of scanning images of four di erent currencies of good
quality and bad quality for training as well as for testing purpose is done.
High resolution scanner is used to acquire the image.

4.2 Image Preprocessing

The acquired image is rst converted into grayscale and then into black and white
image In this step Median lter is used for de-noising the image after resizing the image.

4.2.1 Median Filter

It considers each pixel at the same time and looks at its neighboring pixels to
decide whether it is representative of its surroundings or not.
The pixel value is replaced with median of neighboring pixel value
Median lter is more e ective when its goal is to simultaneously reduce noise and
at the same time to preserve edges in the image
From the input color image each Red, Blue and green channels are denoised
separately by using median lter.
8
 CHAPTER 4. PROPOSED APPROACH

Fig 2 : Noiseless image using Median Filter

4.3 Feature Extraction

The features extracted constitute input vector to the system. The potential
fea-tures in the image based on color, texture and shape are extracted for e cient
classi - cation. The input vector can be written as follows:

F= (cm, cv, cs, eh, glcm ,e)

Euler number of an image is a scalar value which represents the number of objects
in the image minus the total number of holes in those objects

X
e = Euler = O Ho

where O stands for any object in the image, and Ho stands for any hole in that
object. The mean color intensity cm, then color variance cv, nally color skewness cs
are com-puted for each channel and is stored in input feature vector.
9
 CHAPTER 4. PROPOSED APPROACH

Table 1 : Parameters of input vector for Feature Extraction

Initially the color image is transformed into LUV color space

4.3.1 LUV Color Space

In colorimetry, the CIE 1976 (L*, u*, v*) color space, commonly known by its
abbreviation CIELUV, is a color space adopted by the International Commission on
Illumination (CIE) in 1976, as a simple-to-compute transformation of the 1931 CIE
XYZ color space, but which attempted perceptual uniformity. It is extensively used
for applications such as computer graphics which deal with colored lights. Although
additive mixtures of di erent colored lights will fall on a line in CIELUV's uniform
chromaticity diagram (dubbed the CIE 1976 UCS), such additive mixtures will not,
contrary to popular belief, fall along a line in the CIELUV color space unless the
mixtures are constant in lightness.
10
 CHAPTER 4. PROPOSED APPROACH

Fig 3 : LUV Color Model

4.4 Shape Feature

4.4.1 Edge Detection

Edge detection is a fundamental tool in image processing particularly in the
area of feature extraction which captures sharp changes in the image brightness.
Edge de-tection re ects the sharp intensity changes in the colors of the image.It
identi es object boundaries in the image. There are so many algorithms which
include Sobel, Prewitt, Roberts and Canny. Canny is powerful because it detects
weak edges and it has low error rate.A 1x5 edge orientation histogram is computed
(horizontal, vertical,2 diagonals and 1 non-directional) eh is computed. A sobel mask
is applied to nd the max sobel gradient and index of orientation then the Canny edge
detection is used for nding edges in the image. The edge orientation histogram with
ve bins one for each edge type is computed and is stored in the input feature vector.
11
 CHAPTER 4. PROPOSED APPROACH

4.4.2 Gray level co-occurrence matrix

Each currency note is characterized by its texture. Therefore the texture feature
is considered as an important feature to discriminate currency notes having same color.
Gray level co-occurrence matrix (GLCM) is used for analyzing the
texture. It is a statistical method of examining the texture
It considers the spatial relationship of the pixels.
It characterizes the texture of the image by calculating how often pairs of
pixels with speci c values and in speci ed area occur in the image.

After creating a GLCM, some statistical measures are extracted from the matrix
as:
1.Contrast : Returns a measure of the intensity contrast between a pixel and its
neighbor over the whole image.
2
Range = [0 (size(GLCM,1)-1) ]
Contrast is 0 for a constant image. The property Contrast is also known as
variance and inertia.
2.Correlation : Returns a measure of how correlated a pixel is to its neighbor over
the whole image.
Range = [-1 1]
Correlation is 1 or -1 for a perfectly positively or negatively correlated image.
Corre-lation is NaN for a constant image
3.Energy : Returns the sum of squared elements in the GLCM.
Range = [0 1]
Energy is 1 for a constant image.
The property Energy is also known as uniformity, uniformity of energy, and
angular second moment.
4.Homogeneity : Returns a value that measures the closeness of the distribution
of elements in the GLCM to the GLCM diagonal.
Range = [0 1]
Homogeneity is 1 for a diagonal GLCM.
12
Chapter 5

Classi cation Using Arti cial Neural

Network

Once the features are extracted it is essential to recognize the currency based on
these features by using an e ective currency recognition system called classi er. One of
the most common classi er which had been used recently is Arti cial neural network.

Fig 4 : Layers in Arti cial Neural Network

13
 CHAPTER 5. CLASSIFICATION USING ARTIFICIAL NEURAL NETWORK

5.0.1 Training Phase

Initially the feedforward backpropagation neural network is trained with 100
sam-ples of each currency type.Feed forward backpropagation neural network
originate from the problem of interpolating a set of data in a hyperspace to nd the
best hyper-plane that interpolates that set of data.
In feedforward backpropagation Neural Networks, there are three
layers: Input layer, Hidden layer and Output layer.
Input Layer
The input layer consists of the data (or the pattern) that is to be interpolated (or
classi ed).
Hidden Layer
The hidden layer provides a nonlinear transformation from the input layer space
to the hidden layer space. It usually consists of a high number of neurons.
Output Layer
The output layer provides a linear transformation from the hidden layer space to
the output layer space.

5.0.2 Testing Phase

Once the network has been trained on di erent currencies with the their corre-
sponding feature vectors, along with target class and the trained network is validated
against a new test image to check the performance of the proposed system .
The Experimental results are as follows :
14
 CHAPTER 5. CLASSIFICATION USING ARTIFICIAL NEURAL NETWORK

Fig 5 : Performance of Network Classi er

Epoch :

Presentation of the set of training (input and/or target) vectors to a network and
the calculation of new weights and biases.Training vectors can be presented
one at a time or all together in a batch.

Best validation performance is 0.0018166 at epoch 0

15
 CHAPTER 5. CLASSIFICATION USING ARTIFICIAL NEURAL NETWORK

Fig 6 : Training State of Network Classi er

The recognition results are as shown in Table 2. Average recognition rate was
seen as 93.84 which is quite reasonable and acceptable in various cases.

Table 2 : Experimental Results of feedforward back propagation Network Classi er

16
Chapter 6

Conclusion and Future Scope

Paper Currency Recognition is an important application of Pattern

Recognition. Many studies were made to recognize paper currencies using Neural
Networks. In this paper, another method of recognizing currencies has been
introduced, which is based on correlation between images. The method uses feed
forward backpropagation neural network. The method is quite reasonable in terms of
accuracy. However, there is a room to improve the processing time.
The proposed algorithm is fully automatic and requires no human intervention.
The author is also thinking to apply the proposed feature methodology for another
model of classi cation. It might improve the accuracy and e ciency process. This
work is in progress as a subsequent work together with the issue of considering
dimensionality reduction with genetic algorithms.
17
References

[1] Ch.Ratna Jyothi, Dr.Y.K.SundaraKrishna, Dr.V.SrinivasaRao. Paper Currency

recognition for color images based on Arti cial Neural Network. International Con-
ference on Electrical, Electronics, and Optimization Techniques (ICEEOT) - 2016

[2] Ayush Arora. Currency recognition system using canny edge detection.Journal
of technological advances and scienti c research(JTASR).

[3] Hassanpour H, Farahabadi, PM. Using Hidden Markov Models for paper
currency recognition, Expert Systems with Applications 2009; 36(6):

[4] Gonzalez RC, Woods RE. Digital image processing, Prentice Hall; 2009

[5] Hassanpour H, Mesbah M. Neonatal EEG seizure detection using spike signatures
in the time-frequency domain, IJE Transactions A: Basics 2007; 20(2):137-146.

[6] Hassanpour H, Yaseri A, Ardeshir G. Feature Extraction for Paper currency

Recog-nition. In International symposium on signal processing and its
applications (ISSPA), Sharjah, UAE; 2007.

[7] Iosifescu M. Finite Markov process and their applications. New York, NY: Wiley;
1980.

[8] Jae LS. Two-dimensional signal and image processing. Englewood Cli s, NJ:
Pren-tice Hall; 1990.
18
 REFERENCES

[9] Takeda F, Nishikage T. Multiple kinds of paper currency recognition using

neural network and application for Euro currency. In IEEE International Joint
Conference on Neural Networks, Como, p. 143147; 2000.

[10] Zhang EH, Jiang B, Duan JH, Bian ZZ. Research on paper currency recognition
by neural networks. In Proceeding of the second international conference on
machine learning and cybernetics, p. 2193 2197; 200.
19