Vous êtes sur la page 1sur 6

Recognition System for

Pakistani Paper Currency


Todays era is termed as the IT age or we
can say
it Computers Age so it is thought
inevitable to have software solutions for
various problems in order to save
the time, in different organizations despite
of it whether the organization is small or
large enough. We focus to In order
recognize Pakistani paper currency
accurately.
The Pakistan paper currency has different
denominations, with each of them looking
totally different. For instance the size of the
paper is different, the same as the color and
pattern. Otherwise our system is based on
image processing, techniques which include
Noise removal, preprocessing, In order to
make the system more comprehensive, we
need to create a small database for storing
the features of the currency. The system will
be programmed based on MATLAB and
include a user-friendly interface. The main
steps in the system are:
Read image, reading the image we get
from the scanner as well as the format of
the image is JPEG.

We have scanned 100 Pakistani paper


currency notes
from different currency domain (Rs. 10, 20,
50, 100, 1000). Each note was scanned at
200 dpi using an HP scanner. The system is
based on scanner, PC and classifiers. Here
the classifier that we have used is Knn
classifier that is in fact a nearest
neighborhood classifier. Before applying a
classifier we had used preprocessing
techniques, in such a way that we had first
converted the RGB image to grey scale
image, in order to remove noise we had
used wiener filter. The software will extract
the features of the test images. Once these
five features will be extracted they will be
matched with the features stored in MAT
file. The features in MAT file are features of
train images. If the features of test image
will be matched with the features in MAT file
the software will return the class of that
currency note. If the test image features
dont with any of the features in Mat file the
software display that it is doesnt belong to
any class. the algorithm which we had used
for classification is knn classify. Instancebased classifiers such as the knn classifier
operate on the premises that classification
of unknown instances can be done by
relating the unknown to the known
according to some distance/similarity
function.

Preprocessing, removing noise.


Feature extraction, classification.
Result.
There are 7 denominations of Pakistani
paper currency. Each note has different size
and different color. This system will extract
the features of the test image and will
match with the features stored in training
database (mat file). If the features match it
will display the type of currency.
MATERIALS AND METHODS

System Architecture
Acquisition of Images:

We had scanned 100 Pakistani notes of


different dominations. Each note scanned at
200 dpi using an HP scanner.
Preprocessing:
the preprocessing techniques used are
conversion of RGB image to Gray, removal
of Noise and conversion to Binary.
Conversion to Gray
Noise Removing:
MATLAB includes many noises and removing
filters. The filter which we used is wiener2.
Conversion to Binary:
It assigns a 0 for black and a 1 for white.
Feature Extraction:
feature extraction is a special form of
dimensionality reduction. The features of
the image which we had extracted are Euler
Number, Area, Height, Width, Aspect ratio of
that image.
Classification:
The classification is the process to classify
the currency note into its correct class.

Eroded Money Notes


Recognition using Wavelet
Transform
As a result, Iranian money notes have half
lifetime. Since obliterating such a huge
amount of money and
republishing them impose great economical
burden to the country, eroded money notes
remain in the business cycle for longer
period. Banks and charity organizations
mostly deal with such kind of money notes.
Figure 1 shows some parts of eroded money
notes. In this paper we proposed a new
method for eroded money notes recognition
which is based on texture information and
wavelet transform. To increase classification
accuracy, our system utilizes template
matching technique and courtesy amount of
the money note.

Fig.1.Some samples of eroded money notes.


The proposed money recognition
method
Block diagram of the proposed algorithm is
shown in
Figure 2. First, obverse of the money note is
separated from its reverse side. Then,
region of interest including texture is
extracted from central part of the money.
Using wavelet analysis, some features are
obtained. After feature extraction, the input
image is classified into one of classes,

comparing with the images in the database.


Post-processing step is also performed to
increase recognition accuracy using
courtesy amount and template matching
technique.

Emam Khomaini Using a face detection


algorithm , and having prior knowledge
about the face position, obverse and reverse
sides of a note can be differentiated.
ROI selection
Regions with fine texture have more
information compared to coarse one. In this
paper, different parts of a note are
investigated to find region of interest (ROI),
leading the best performance. Using ROI
instead of the whole money note image not
only improves classification accuracy but
also increases recognition time. Figure 4
shows some regions of a money note with
different objects and textures.

Fig.2.Block diagram of the proposed system.


Database
Usually in money notes recognition
application number of classes do not exceed
10 groups. Iranian money notes can be
considered as 6 major classes (1000, 2000,
5000, 10000, 20000, and 50000 Rials). 20
eroded samples are gathered (total of 120
samples). A money note is called eroded if
an automated teller machine (ATM) can not
handle it. These include obliterated, torn,
patched, altered, and erased, money notes
(Fig.1). Since the proposed system is
expected to differentiate obverse of the
money note from its reverse side, both side
of each money note are scanned in graylevel format with resolution of 300dpi (total
of 240 images).
Obverse from reverse separation
Obverse of all Iranian money notes contain
a picture of

Fig.4. result of face detection algorithm.


Feature extraction
Discrete wavelet transform (DWT), applied
for discrete signals, can be easily performed
by digital computers. DWT has high speed
and needs low memory capacity. In wavelet
domain, signals can be locally characterized
in both time domain and frequency domain
simultaneously. In DWT, scaling and shifting
are performed by digital filters. The input
signal is passed through different filters with
different cut-off frequencies and different
scaling factors. High pass and low pass
filters are used frequently In this paper, we
used two types of features extracted by
packet wavelet analysis (PWA), and
maximum overlap algorithm (MOA).
Packet wavelet analysis (PWA)

In this method, low-pass and high-pass


filters are used to decompose the input
image into sub-bands.
Maximum overlap algorithm (MOA)
In this method, we filter the original image
by the high pass filter H and low pass filter
G to produce the four bands LL, LH, HL and
HH (without any subsampling)
Distance measurement
Some different distance criteria such as
Euclidean (1),
Cosine (2), and Mean square error (MSE) (3)
distances have been used in this paper to
measure the distance between ROI in the
input image and their corresponding in the
database.
Post-processing
Courtesy amount of each money note is
written in both
Latin and Farsi formats at its corner.
Template matching technique and courtesy
information are used to increase
classification rate. We used texture
information rather than courtesy information
in this paper. Because courtesy amount is
fragile and in the most eroded money notes
they are obliterated.
Experimental results
To evaluate the proposed method, we used
240 images of scanned eroded Iranian
money notes. Since the proposed method is
based on the texture of obverse sides,
obverse from reverse separation step

affects the overall performance. In the last


experiment, performance of wavelet based
feature extraction methods (MOA and PWA)
was investigated. Euclidian distance
measure and PWA outperforms the others
with 83% accuracy.

Paper
Recognition
System for
Pakistani Paper
Currency
Eroded Money
Notes
Recognition
using Wavelet
Transform
A Novel Paper
Currency
Recognition
using Fourier
Mellin Transform,
Hidden Markov
Model and
Support Vector
Machine

Yea
r
20
13

currency

Paper
image
Whole
paper

technique

accura

Features
extraction

100%

20
10

Farsi
money
notes

Whole
paper

Template
matching

83%

20
13

150
banknotes
of 101
different
denominati
ons, from
23
countries

Intact
examples

FourierMellin
transform
and SVM

98.7%

Pakistani
Paper
Currency

Vous aimerez peut-être aussi