Al-Nahrain Journal of Science Vol.22 (2), June, 2019, pp.

78-86

An Image Processor Bill Acceptor for Iraqi Currency

Ammar Awni Abbas
Computer Center, University of Baghdad, Baghdad-Iraq.
Corresponding Author :[email protected].
Abstract
Automatic Teller Machines (ATMs) and vending machines is a rapidly growing trade in the
Iraqi market. Bill acceptors\validators, are considered the heart of these machines have a great
significance. Due to the inflation in Iraq with numerous forgery attempt to the currency, an efficient
way to accept and validate the Iraqi currency is very important. This paper suggests a bill
acceptor\validator to accept, recognize the amount, and validate the bill based solely on image
processing functions rather than the magnetic field detection that is used in the regular Bill
Acceptors. The program couples the faked bill with a photo graph of the person who presents it.
The program uses the Matlab computer vision abilities to recognize the bill. The distinctive
characteristics that appears on the banknote when an ultraviolet light is shed on the bill it used for
the authentication part with the same Matlab pattern recognition functions. The proposed system
include all the mechanical and electronical parts that are used in the proposed system. The results
was 100% accurate, which implies that there is great possibility to use this technique in industrial
bill acceptors. [DOI: 10.22401/ANJS.22.2.10]

Keywords: Bill Acceptors, Validators, Matlab, currency authentication, Image Processing.

1. Introduction results for the operation, the sixth section
The Iraqi currency system has been discusses results and main possible
renewed completely after the 2003 events, new enhancement for the suggested apparatus; the
bills with a very high security standards was seventh section is the references.
pumped into the circulation in an
unprecedented way. Despite the high security 2. Related Work
measures that has been taken to prevent A portable X-ray fluorescence (pXRF)
forgery, the Iraqi currency was subjected to spectrometer is used to detect the Peruvian
numerous forgery attacks specially the high banknotes by comparing the spectra of original
amount bills. This results in great disturbance and fake notes, Peruvian Nuevo Sol banknotes
in the markets and consequently in the society were studied: 4 genuine and 3 fake ones, in
in general. While the brightness and the 11 points of analysis for each one, with 77 data
texture characteristics of the note may fade set. The correlation analysis of spectra among
away with time, other characteristics like original notes was 1.0, and there was no
Identification mark, security thread, latent correlation with fake banknotes [1]. The most
image and watermarks should be persistent frequent related work is dedicated to currency
with time and not affected with degradation recognition using image processing and the
[4]. This paper targets an authentication visual characteristics of the note as [2]. A
feature in the Iraqi currency that is ultraviolet combined method between Principal
illumination. An integrated ultraviolet Component Analysis (PCA) and back
fluorescent fiber which creates a very special propagation neural network (BPNN) is
shapes and colors when the note is exposed to proposed by [3] to detect banknotes, for the 24
an ultraviolet source of light. This property is contrast experiments, the training set
used in this paper as the main authentication composed of the banknote images captured by
characteristics. The second section of the webcam results in better recognition results
paper reviews the work of other authors, the than the training set made up of scanned
third section outlines the proposed system banknote images, the highest average
(both hardware and software), and the fourth F-measure is 0.9980 obtained by combination
section discusses the way the apparatus No.9 (Color+LBP_PCA_BPNN) when using
process each bill, the fifth section lists the the banknote images captured by webcam for

78
 Ammar Awni Abbas

training. In [4] a complete survey for all fields universal recognition program for all
of currency detection and authentication plus currencies is the main objective for that paper,
recognition of the serial number of the the proposed algorithm was tested on five
banknote is presented, Techniques used in currencies with huge success. The process of
each of the four areas recognize banknote recognition using Matlab with the most
information (denomination, serial number, common security characteristics is discussed
authenticity, and physical condition) based on in [13]. An embedded system for detecting the
image or sensor data, and are actually applied fake currency is proposed by [14], which
to banknote processing machines. A scale- relied on some special currency characteristics,
invariant feature transform (SIFT) algorithm a neural network is used for recognition
combined with mobile camera is used to and a Light Emitting Diode (LED) for
recognize the currency [5], the correct authentication. The success rate was 100%. In
recognition average for the color SIFT is 0.79 [15] the shape and color information is used in
While, for Gray SIFT the correct recognition the algorithm, two currencies are used in the
average is 0.53. Similarity mapping used to data base of the program. The results showed
recognize banknotes images captured by light that the scanned images have better results
sensor in [6], the proposed method has an error than images taken by a digital camera. In [16]
rate of 0.002 with rejection rate of 0.004 which an ensemble neural network (ENN) is used, by
is much better than other methods. In paper using negative correlation learning (NCL) the
[7], the authors tried to use simple algorithms system gains experience on different parts or
to increase the speed of recognition for the portion of input patterns in an ensemble. After
Chinese yuan (RBM) with great efficiency, the applying certain filters to classify the acquired
recognition ratio of the (100, 50 old style, 50 image a neural networks method for
new style, 20 and 10 of the RBM bill) is (95%, identification is used by [17]. [18] Used the
99%, 99%, 92% and 98%) respectively. In [8] mobile phone camera to make the recognition
an image processing technique is used for the in real time to help people with impaired
recognition process in Matlab program, the visual conditions.
authentication of the 100 teka (currency of
Bangladesh) is made with 100% success rate. 3. Instrument and software setting
[9] Proposed a radial basis for classification The proposed system is shown in Fig.(1),
that uses basic features with correlation to it's composed of the following Hardware parts:
detect the currency, the overall Average 1- Motion sensor, to detect the currency is
Recognition Rate for the data of 110 images is being presented by the user.
computed as 91.51%. The proposed algorithm 2- Servo motor to roll the paper in, when it
is fully automatic and requires no human gets the signal from the motion sensor.
intervention. The characters of the note is 3- Plastic clear Chamber, which consists of
heuristically analyzed to get the exact features two layers of plastic plates joined together
of them before the recognition in [10], by glue forming a clear chamber to host the
Heuristic analysis of Serial Number for 9 rolled in note, the servo motor (1) roll in the
character, taking in account four individual banknote and let it lay on this chamber for
properties of a character(hue, saturation, further processing.
contrast and brightness) reveals that particular 4- Web camera (1) for the recognition and
trend in terms of values of hue and saturation, authentication, this camera faces the plastic
contrast and brightness. It can be concluded chamber from horizontal view.
that while contrast and brightness are 5- Arduino Uno microcontroller to connect to
dependent on the extrinsic factors of sensor, the relay.
hue and saturation are more likely dependent 6- Two-Channel relay: connected to the
on the consistent pattern design of note. In Arduino Uno which is used to
[11] a review of the recognition processes is activate/deactivate the White and
made. After applying certain filters to classify Ultraviolet lights.
the acquired image a neural networks method 7- White bulb light: used for recognition
for identification is used by [12], making a process.

79
Al-Nahrain Journal of Science Vol.22 (2), June, 2019, pp.78-86

8- Ultraviolet light: used for the authentication

process.
9- Servo motor (2): used to roll the paper out
of the plastic chamber when the
recognition/authentication processes is
over, the note is guided to the currency
accumulation sack.
10- Web camera (2): to take a photograph for
the Fake currency provider.

All of these components are encapsulated

in well isolated box to prevent the effects of
any unwanted lights. As shown in Fig.(1).

Fig.(1): The proposed Bill Acceptor Hardware.

The apparatus is programmed using

Matlab programming language, with the
following additional packages installed:
1. Matlab Support Package for USB Web
cams.
2. Matlab Support Package for Arduino
Hardware (Arduino microcontroller
can be replaced by the Raspberry Pi
microcontroller).
The following Matlab toolboxes is used
and must be installed:
1. Computer Vision System Toolbox.
2. Image Acquisition Toolbox.
3. Image Processing Toolbox.
A flowchart of the proposed bill acceptor
operation is shown in Fig.(2).

80
 Ammar Awni Abbas

Fig.(2): A Flowchart of the Bill Acceptor operation.

4. Test the sensor sends a signal to a servo motor

A bill acceptors have one basic purpose is to operate, the motor attached to serrated
to accept and validate paper currency in roller, this event suck the bill inside the
exchange for a product or service. The bill machine, another signal is sent to trigger
acceptor "grabs" the bill, scans it for validity, the start of the Matlab program. The note
and sends a corresponding Input/Output signal is guided to the clear plastic chamber
to its output connector to record the through a proper plastic guide; this
transaction. The proposed device process the chamber serve as temporary station,
bill as follows: where the recognition and the
1- The bill presented to the proposed device authentication is made.
is sensed by motion sensor, that in turn,
81
Al-Nahrain Journal of Science Vol.22 (2), June, 2019, pp.78-86

2- The system is connected via Arduino Uno detect the amount of the banknote. The process
microcontroller to a two channel-relay of currency recognition consists of five steps
that switch ON an ordinary white light, that is repeated between the target image and
followed by, a taking a photo in shape of each of the templates:
frame captured from the live feed of web 1- Read both the Template images and the
camera (1) attached to the system, the target image.
photo is used to recognize the banknote 2- Detect feature point of both images using
amount by applying the pattern the detectSURFFeatures function, this
recognition software between the target function returns a Speeded-Up Robust
image and a set of templates that is a Features (SURF) Points object, points,
scanned image of the various containing information about SURF
denomination of notes. features detected in the 2-D grayscale input
3- The program send a signal to the Arduino image I. The detectSURFFeatures function
Uno microcontroller, which in turn sends implements the Speeded-Up Robust
a signal to the two channel relay to turn Features (SURF) algorithm to find blob
ON the Ultraviolet light. features.
4- Another photo is taken from the web 3- Extract feature descriptors: using
camera (1) is taken when the Ultraviolet extractFeatures function which Extract
light is ON, and the same recognition interest point descriptor. The function
program is activated with another set of derives the descriptors from pixels
templates that is made with Ultraviolet surrounding an interest point. The pixels
light. This photo is used to authenticate represent and match features specified by a
the currency. single-point location. Each single-point
5- If the currency is authentic, a servo motor specifies the center location of a
(2) is turned ON to push the currency to neighborhood. The method used for
the currency accumulation sack. With the descriptor extraction depends on the class
value registered in the system for further of the input points, in this case we used
operations. Speeded-Up Robust Features (SURF)
6- Otherwise, the servo motor(2) also turned obtained from the second stage.
ON to push the currency to the 4- Find the putative point matches: using the
accumulation sack with a flag raised to matchFeatures function which match the
program to take a photo of the person who features using their descriptors. This
present the fake banknote using web function returns indices of the matching
camera (2). This will prevent fake features in the two input feature sets
currency from circulating in the monetary obtained from Templates and target
system. images. The input feature is a binary
The recognition process is made by Features image.
exploiting the capabilities of the computer 5- The program detects the matched features
vision toolbox in MATLAB. By discovering between the target and each of the
point correspondences between the reference template, the image with the largest match
templates and the objective picture, the points is considered to be a hit.
matching is made when the template image The same procedure is repeated for the
that has the largest point correspondences with authentication process with the exception that
the target image. the ultraviolet light is turned ON when the
The proposed program is based upon the webcam takes the photo and same applied to
Speeded-Up Robust Features (SURF) which the template images.
uses scale- and rotation-invariant detector and
descriptor. The main idea is to use integral
images for image convolutions.
There are eight denominations in the Iraqi
currency making a total of sixteen template
pictures that the program must choose from to

82
 Ammar Awni Abbas

5. Experiments and Results

The experimental stage consists of
recognizing and authenticating a set of 2400
banknotes, 300 banknotes per denomination
class. The notes are acquired from the ordinary
monetary circulation where the bills are
sometimes partially torn or handwritten notes.
The results of the recognition part is shown in
Figs. [3, 4, 5, 6and 7]. Table (1) shows the
results of the testing.

Table (1)
The Recognition and Authentication results.
50 250 500 1000 5000 10000 25000 50000
Banknote
Dinars Dinars Dinars Dinars Dinars Dinars Dinars Dinars
No. of authentic
290 290 290 290 290 290 290 290
notes tested
No. of forged notes
10 10 10 10 10 10 10 10
tested
Total 300 300 300 300 300 300 300 300
Recognition 300 300 300 300 300 300 300 300
Authentication 300 300 300 300 300 300 300 300
Error 0 0 0 0 0 0 0 0
Accuracy 100% 100% 100% 100% 100% 100% 100% 100%

Fig.(3): Matching of 50 Dinar denomination Fig.(4): Matching of 500 Dinar

(Back). denomination (Front).

In Fig (3) the target image on the left that The target image on the left and the best
is processed by the suggested device and the match coupled with the target on the right for
match is found, for descriptive reasons the the 500 dinar denomination.
target and the best match are presented in the
same image on the right with the match points
on both images illustrated on both images.

83
Al-Nahrain Journal of Science Vol.22 (2), June, 2019, pp.78-86

Fig.(8): Matching of 50000 Dinar

Fig.(5): Matching of 10000 Dinar denomination (2-Sides of the Note).
denomination (Front).
In Fig.(8) we scanned two side of the bill
as one target image and again the program
succeed in finding the best match.
The accuracy of the results is 100%, the
pattern matching method used is both concise
and precise. The clustering of instructions in
the form of functions in MATLAB creates the
ability to get the maximum functionality
with a minimum amount of lines of code.
MATLAB support for microcontrollers,
extended the capabilities of MATLAB to a
whole new level that can be used to create a
Fig.(6): Matching of 25000 Dinar very wide mechatronics applications including
denomination (In Cluttered Scene). the proposed bill acceptor. Computer vision
toolbox made the recognition process easy,
The target image on the right can be fast and accurate. The target image can be
detected even if the best match found in a identified even if the note is placed in a
scene that contains both faces of the currency cluttered scene or if the target is rotated or
as shown in Fig.(6). partially torn. As a result, there is no need for
preprocessing, image registration, or any other
enhancement methods. For the authentication
part only the part that contains the ultraviolet
sensitive fiber is used for the pattern matching,
this part is cropped from the templates
(exposed to ultraviolet light), as shown in
Fig.(9).

Fig.(7): Matching of 25000 Dinar

denomination (In Complex Cluttered Scene).
In Fig.(7) the match for the target image no
matter how complex the scene is complicated
and no matter the difference in the size of the
target and the templates. Fig.(9): Cropping the Ultraviolet part
of the note.

84
 Ammar Awni Abbas

The region of interest ROI in case of Raspberry Pi Camera Module v2). Both
authentication is the phosphoric rectangle that microcontrollers are supported by Matlab and
appears after shedding the ultraviolet light, can benefit from the computer vision
cropping that part, speeds up the process of capabilities of Matlab. The algorithm is very
authentication and eliminates the use of robust, it can distinguish images in spite of a
unnecessary preprocessing operation. The scale change, rotation, background color or
program couples the currency denomination impediment. The computer vision functions
that is used as target for the recognition minimize the amount of coding needed while
processing with a corresponding phosphoric retaining high speed, so there is a possibility of
rectangle for that certain currency to be used operation in real time. To further enhance the
for the authentication process. work by taking the following measures:
1.Adding more security features to be
checked by the algorithm such as security
thread and water marks.
2.The ability to recognize and authenticate
currencies other than the Iraqi currency, by
expanding the algorithm to accept other
banknotes templates.

7. References
[1] Jara M., Obregón C., Del Castillo C.
"Exploratory analysis for the identification
of false banknotes using portable X-ray
Fluorescence spectrometer", Applied
Radiation and Isotopes, 135(1), 212–218,
Fig.(10): The Note is authentic, a match 2018.
found between target and template. [2] Rajebhosale S., Gujarathi D., Nikam S.,
Gogte P., Bahiram N., "Currency
Recognition System Using Image
Processing", International Research
Journal of Engineering and Technology
(IRJET), 4(3), 2599-2601, 2018.
[3] Ren Y., "Banknotes Recognition in real
time using ANN", M.Sc. thesis presented
to the university of Auckland, New
Zeeland, 2017.
[4] Lee J., Hong H., Kim K., Park K., "A
Survey on Banknote Recognition Methods
by Various Sensors", Sensors Journal,
17(2), 235-359, 2017.
Fig.(11): The Note is forged, No match found [5] Abu Doush L., AL-Btoush S., "Currency
between target and template. recognition using a smartphone:
Comparison between color SIFT and gray
The algorithm will find no match for the scale SIFT algorithms", Journal of King
phosphoric rectangle in the forged note as Saud University- Computer and
shown in Fig.(11). Information Sciences, 29(4), 484-492,
2017.
6. Discussions
[6] Pham T., Park H., Kwon S., Park K.,
The suggested configuration can be turned
Jeong D., Yoon S., "Efficient Banknote
into a standalone device. The algorithm can be
Recognition Based on Selection of
uploaded to the microcontroller such as
Discriminative Regions with One-
Arduino or Raspberry Pi where the webcams
can be replaced by (OV 7670for Arduino or

85
Al-Nahrain Journal of Science Vol.22 (2), June, 2019, pp.78-86

Dimensional Visible-Light Line Sensor", Network Ensemble", JOURNAL OF

Sensors, 16(3), 328-346, 2016. MULTIMEDIA, 5(6), 560-567, 2010.
[7] Bhurke C., Sirdeshmukh M., Kanitkar [17] Hassanpour H., Farahabadi P., "Using
M.S., "Currency Recognition Using Image Hidden Markov Models for paper currency
Processing", International Journal of recognition", Expert Systems with
Innovative Research in Computer and Applications, 36(6), 10105-10111, 2009.
Communication Engineering, 3(5), 4418- [18] Liu X., Doermann D., "A Camera Phone
4422, 2015. Based Currency Reader for the Visually
[8] Saifullah S. M., Ananna A.R., Hossain S., Impaired", ASSETS’08, ACM 978-1-
Hossain J., Zishan S.R., "Currency 59593-976-0/08/10, 2008.
Recognition System Using Image [19] Zhang Er., Jiang Bo., Uan Jh., Bian Z,
Processing", American Journal of "Research on paper currency recognition
Engineering Research, 4(11), 26-32, 2015. by neural networks", Proceedings of the
[9] Sarfraz M., "An intelligent paper currency Second International Conference on
recognition system", Procedia Computer Machine Learning and Cybernetics, Xian,
Science, 65(1), 538 – 545, 2015. 1(1), 2193-2197, 2003.
[10] Singh Reel P., Krishan G., Kotwal S., [20] Bay H., Tuytelaars T., Van Gool L.,
"Image Processing based Heuristic "SURF: Speeded Up Robust Features", In:
Analysis for Enhanced Currency Leonardis A., Bischof H., Pinz A. (eds)
Recognition", International Journal of Computer Vision – ECCV 2006, Lecture
Advancements in Technology, 2(1), 82-89, Notes in Computer Science, 3951(1), 346-
2014. 359, 2006.
[11] Shirsath A, Bharkad S, "Survey of
Currency Recognition System Using
Image Processing", International Journal of
Computational Engineering Research,
3(7), 36-40, 2013.
[12] Jain V., Vijay R., "Indian Currency
Denomination Identification Using Image
Processing Technique", International
Journal of Computer Science and
Information Technologies, 4(1), 126 – 128,
2013.
[13] Mirza R., Nanda V., "Paper Currency
Verification System Based on
Characteristic Extraction Using Image
Processing", International Journal of
Engineering and Advanced Technology
(IJEAT), 1(3), 68-71, 2012.
[14] Pathrabe T., Karmore S., "A Novel
Approach of Embedded System for Indian
Paper Currency Recognition",
International Journal of Computer Trends
and Technology- 4(8), 152-156, 2011.
[15] Wang Y, Ling s, "Currency recognition
system using image processing", Bachelor
Thesis, Akademin för teknik och miljö
Högskolan i Gävle S-801 76 Gävle,
Sweden, 2010.
[16] Debnath K., Ahmed S., Shahjahan M.,
"A Paper Currency Recognition System
Using Negatively Correlated Neural

86