2023 7th International Conference On Computing, Communication, Control And Automation (ICCUBEA)

Pimpri Chinchwad College of Engineering (PCCOE), Pune, India. Aug 18-19, 2023

Fake Currency Detection using Image Processing

Sagar Shinde Lalitkumar Wadhwa Shreenand Naik
2023 7th International Conference On Computing, Communication, Control And Automation (ICCUBEA) | 979-8-3503-0426-8/23/$31.00 ©2023 IEEE | DOI: 10.1109/ICCUBEA58933.2023.10392045

Electronics & Telecommunication Electronics & Telecommunication Electronics & Telecommunication

Dr. D. Y. Patil Institute of Technology Dr. D. Y. Patil Institute of Technology Dr. D. Y. Patil Institute of Technology
Pune, India Pune, India Pune, India
[email protected]

Rohan Kudale Nitin Sherje Karan Mohnani

Electronics & Telecommunication Mechanical Engineering Electronics & Telecommunication
Dr. D. Y. Patil Institute of Technology Dr. D. Y. Patil Institute of Technology Dr. D. Y. Patil Institute of Technology
Pune, India Pune, India Pune, India

Abstract—Advancements in color printing have led to a rise have been advocating for the development of a reliable and
in the number of counterfeit currency notes being created and effective cash identification system. Nowadays, it is very
duplicated on a massive scale. In the past, it was very difficult common to see machines that can identify banknotes in the
to do this, but now anyone with a laser printer can easily do it.
Due to the rise of fake notes, the number of counterfeit currency vending machines of various types of goods, such as soft
notes has increased significantly in India. This is a major issue drinks, candies, and train tickets. The goal of this system is
that the country is facing due to its various problems such as to extract the necessary information from the currency notes.
black money and corruption. A system that can detect the fake Although there are many methods that are currently being used
currency notes in a fraction of the time is being developed. to identify the notes, the ideal method is to use the multiple
The proposed system will use image processing techniques to
verify the authenticity of Indian currency notes. This method characteristics of the note [11]. For instance, the size and color
will involve extracting various features from the notes. Some of the note are two of the most common factors that are used
of these include the Bleed Line, Water marking, Fluorescence, to determine its authenticity. Unfortunately, this method is not
Security thread, Micro Lettering, and Identification mark. The effective because dirty notes have significantly altered their
proposed system is using software known as MATLAB. The color. In order to improve the accuracy of the system, it must
Support vector machine is a supervised learning model that is
used for extracting the notes’ features. It is also used for analyzing first extract the relevant features from the image.
data related to regression and classification. We have also tested
the other methods related to this proposed work, such as the
graphical black Box algorithm. The proposed system has various II. M ETHODOLOGY
advantages, such as its high-performance speed and simplicity. It
will be able to determine if the currency notes are real or fake. The proposed system as shown in below Fig. 1, can perform
Its accuracy is also measured using the Confusion Matrix.
this task. It uses the captured image of the note under
Index Terms—Image processing, SVM, counterfeiting, fake
currency detection ultraviolet light. Currency notes can be captured by a digital
camera using the ultraviolet light [2]. The captured image is
I. I NTRODUCTION then converted into grayscale using a process known as Edge
Due to the increasing demand for banknote recognition detection. The paper currency’s features are then segmented
systems, researchers have been developing various robust and and cropped. The intensity of the extracted features is then
dependable methods to automate the process. One of the calculated and the note is declared to be real if the condition
most common goals of these systems is to ensure that the is satisfied.
accuracy and speed of the recognition are maintained. Various
applications, such as vending machines and selling goods,
require banknote recognition software [9]. In order to ensure
that the system is reliable and accurate, it must be able to
extract the necessary data from the currency [1]. This is very
challenging for system designers due to the complexity of the
task. The proposed system can be easily implemented using
the Matlab software. The output will determine if the note is
fake. Due to the rapid technological advancements that have
occurred in the financial sector, the need for automatic cash
identification has become more important. This is especially Fig. 1: System Architecture of proposed system.
true in vending machines and sales machines. Researchers

979-8-3503-0426-8/23/$31.00 ©2023 IEEE 1

Authorized licensed use limited to: Marathwada Mitra Mandal's College of Engg. Downloaded on August 22,2024 at 08:47:57 UTC from IEEE Xplore. Restrictions apply.
 A. Image Acquisition
The acquisition of an image involves turning a physical
object into a digital one. For instance, the image shown in
Fig. 2 is of Indian currency notes. It was acquired using a
digital camera and later saved for manipulation.

Fig. 4: Identification marks.

Fig. 2: Image Acquisition.

Fig. 5: Portrait of Mahatma Gandhi.

B. Image pre-processing
In preparation for use in training and inference, image pre-
processing is a process that involves various steps before
data analysis and extraction is carried out. This includes
the preparation of images for scaling, orienting and color
adjustments.
In addition to improving the efficiency of training and
inference, model pre-processing can also help reduce the time
it takes to perform various tasks. For instance, if the input
images are very large, reducing their size can speed up the
training time [3]. Fig. 6: Edge Based Segmentation of Security Thread.

C. Gray scale conversion and edge detection

Due to the information contained in the acquired image, it
has been transformed into a grayscale one. As the processing
of this image continues, the margins will also become visible.

Fig. 7: Edge based Segmentation of serial number.

E. Feature Extraction
Feature extraction is a process that involves reducing the
dimensionality of an image [8]. It allows users to see the
Fig. 3: Gray Scale Conversion.
appealing parts of the image as a compact feature. This method
can be useful when it comes to performing various tasks,
such as searching for and matching pictures [7]. The following
D. Image Segmentation features are being extracted from currency.
In order to process the collected information, segmentation 1) Security Thread: The security thread of the Rs. 2000
is performed. This process involves breaking down the image note is readable and features the inscriptions ”RBI”, ”2000”,
into sections that correspond to objects. and ”Bharat” in Hindi. This design feature indicates that the

2
Authorized licensed use limited to: Marathwada Mitra Mandal's College of Engg. Downloaded on August 22,2024 at 08:47:57 UTC from IEEE Xplore. Restrictions apply.
 banknote’s color changes as it is tilted as shown in below fig.8.
On the other hand, the security thread of the Rs.100 and 500
notes has similar inscriptions [4].
When held up to the light, the security threads of Rs.2000,
Rs.500, and Rs.100 notes can be seen as continuous lines. In
addition, the embedded security threads of Rs.10, Rs.20, and
Rs.50 notes feature the inscriptions ”Bharat” and ”RBI.”

Fig. 11: Mahatma Gandhi Watermark.

5) Identification Mark: Intaglio print can be used by blind

people to identify the denomination of banknotes. For exam-
ple, the denomination of the 2000 notes has seven lines long,
while the 500 has five lines.

Fig. 8: Security thread color shift.

2) Serial Number: The bottom left and top right sides of the
note have a panel that displays the serial numbers of banknotes
as shown in Fig. 9.

Fig. 12: Identification Mark.

F. Image Evaluation
An image analysis is a process that involves breaking
down a picture into its basic elements. This can help retrieve
information such as the type of objects in the image and their
properties.
Fig. 9: Serial Number.
G. Classification Using Support Vector Machine
3) Latent Image: The front side of a currency bill features a After collecting the various features of banknote images,
vertical band that contains a latent image of the denomination it is important that the banknote pattern is identified by the
[12]. It can be seen when the bill is held horizontal as shown extracted characteristics. This paper proposes a method that
in below Fig. 10. uses a classifier to identify the appropriate banknote pattern.
This method is commonly used in support vector machine
analysis [14]. The proposed method can be used to identify
counterfeit banknotes by training the classifier on images that
have been pre-processed. In a working scenario, a banknote
image is taken by a scanner or mobile camera. After going
through the various features of the image, the classifier is able
to identify the genuine notes.

III. S IMULATION R ESULTS

Fig. 10: Latent Image. The image of a currency that has to be verified or checked
as genuine should be included as an input for the system to
4) Mahatma Gandhi Watermark: When exposed to light, process. It can be acquired by using methods such as scanning
the Mahatma Gandhi banknotes feature the watermark with or taking a picture with the phone [10].
multi-directional lines and a shade effect [6] as shown in below The use of a median filter is often regarded as a robust
Fig. 11. technique for reducing noise. It is particularly effective against

3
Authorized licensed use limited to: Marathwada Mitra Mandal's College of Engg. Downloaded on August 22,2024 at 08:47:57 UTC from IEEE Xplore. Restrictions apply.
 Fig. 16: Edge Detection.
Fig. 13: Input Image.

In image processing, a feature extraction program is used

to identify and extract various visual details from an image
[13]. These include shapes, colors, and edges. For instance, if
a feature extraction program is focused on finding the edges
of an image, it would show the detected details.

Fig. 14: Denoised Image.

‘salt-and-pepper’ sounds, and it is strong against the presence

of outliers at the gray-level.

Fig. 17: ROI-Feature Extraction (Texture).

Fig. 15: Gray Image.

Understanding the color image is important when converting

a color image to a grayscale one. A pixel color is usually
represented by the combination of Red, Green, and Blue. A
grayscale image can be layered and viewed as one whole [15]. Fig. 18: ROI-Feature Extraction (Edge & Shape).
There are various techniques that can be used to convert a
colored image to a grayscale one. The flow of results of proposed system has been shown
An image processing technique known as edge detection is in fig. 19 through experimental demonstration whether the
used to find the edges of an object in an image. It can do this currency we’ve analyzed is real or fake.
by detecting the variations in brightness. An algorithm accuracy comparison is a process that involves
A structure or pattern in a raw data set that is recurrent comparing the performance of various algorithms when it
is referred to as statistical. There are various methods that comes to identifying and predicting outcomes for a given
can be used to extract these features, such as transformations, dataset [5]. The objective is to find the one that performs well
structural, and statistical. on the given task, taking into account the various factors that

4
Authorized licensed use limited to: Marathwada Mitra Mandal's College of Engg. Downloaded on August 22,2024 at 08:47:57 UTC from IEEE Xplore. Restrictions apply.
 true positive indicates that the prediction accurately identified
the notes, while the false positive shows that the algorithm
identified the non-100 rupee notes incorrectly.

Fig. 19: Result.

TABLE I: Algorithm Accuracy Comparison. Fig. 22: Graph for Rs. 200 Note Accuracy.

Algorithm Accuracy Precision F-Score A confusion matrix graph for the accuracy of a 200 rupee
Proposed(SVM) 99.9 99.9 99.9 note detection algorithm would display the true positive,
SVC 97.5 99.7 98.6 false positive, true negative, and false negative rates for the
GBC 99.4 99.9 99.7 algorithm’s predictions. The true positive rate represents the
percentage of actual 200-rupee notes that the algorithm cor-
rectly identified as such, while the false positive rate represents
the percentage of non-200-rupee notes that the algorithm
incorrectly identified as 200-rupee notes.
TABLE II: Note Type and their Detection Rate Value

Note Type Detection Rate

100 86.7
200 93.3
500 80
2000 93.3
AVG 88.32 %

Fig. 20: Algorithm Accuracy Comparison.

affect its accuracy, such as the complexity, training data, and

requirements for the application.

Fig. 21: Graph for Rs. 100 Note Accuracy. Fig. 23: Graph For 500 Note Accuracy.

A representation of the accuracy of an algorithm for de- The accuracy of an algorithm that detects the value of 500
tecting 100 rupee notes shows the predicted false negatives, rupee notes is depicted in a matrix with the predicted false
true positives, true positive, and false negative rates. The negatives, true negatives, false positives, and true positive

5
Authorized licensed use limited to: Marathwada Mitra Mandal's College of Engg. Downloaded on August 22,2024 at 08:47:57 UTC from IEEE Xplore. Restrictions apply.
 rates. The false positive indicates that the prediction is able [8] C. Shewale, S. B. Shinde, Y. B. Gurav, R. J. Partil, and S. U. Kadam,
to identify the notes accurately, while the true positive shows “Compiler optimization prediction with new self-improved optimization
model,” Int. J. Adv. Comput. Sci. Appl., vol. 14, no. 2, 2023.
that the algorithm is able to identify the non-500 rupee notes [9] M. N. Shende and P. P. Patil, “A Review on Fake Currency Detection
incorrectly. using Image Processing,” Int J Futur Revolut Comput Sci Commun Eng,
vol. 4, no. 1, pp. 391–393, 2018.
[10] Arya and D. M. Sasikumar, “Fake currency detection,” in 2019 Interna-
tional Conference on Recent Advances in Energy-efficient Computing
and Communication (ICRAECC), 2019.
[11] S. Khoje and S. Shinde, “Evaluation of ripplet transform as a texture
characterization for Iris recognition,” J. Inst. Eng. (India) Ser. B, vol.
104, no. 2, pp. 369–380, 2023.
[12] K. Sawant and C. More, “Currency recognition using image processing
and minimum distance classifier technique,” Int. J. Adv. Eng. Res. Sci.,
vol. 3, no. 9, pp. 1–8, 2016.
[13] K. B. Zende, B. Kokare, S. Pise, and P. S. Togrikar, “Fake Note Detection
System,” IJIRT, vol. 4, pp. 46–49, 2017.
[14] S. B. Shinde, M. Alagirisamy, K. Wagh, and P. Dhore, “Math accessi-
bility for blind people in society using machine learning,” ECS Trans.,
vol. 107, no. 1, pp. 18071–18090, 2022.
[15] A. Ali and M. Manzoor, “Recognition system for Pakistani paper cur-
rency,” Res. J. Appl. Sci. Eng. Technol., vol. 6, no. 16, pp. 3078–3085,
2013.
Fig. 24: Graph For 2000 Note Accuracy.

A representation of the accuracy of a currency detection

algorithm that predicts the value of 2000 rupee notes shows
the true, false, true negative, and false positive rates. The true
positive indicates that the algorithm has correctly identified
the actual notes, while the false positive shows that it has
mistakenly identified the non-2000-rupee notes.
According to the features mentioned above, got 93.3%
correct output as a accuracy.
IV. C ONCLUSIONS
The algorithm for detecting fake money was developed
on the basis of image processing in MATLAB. It takes into
account various features of currency notes, such as the security
thread, serial number, and the Mahatma Gandhi portrait etc.
The algorithm is able to perform well with the new 500 and
2000 rupee notes. The confusion matrix shows 93.3% accuracy
R EFERENCES
[1] I. Santhiya, “Research on Fake Indian Currency Note Detection using
Image Processing,” International Journal of Scientific Development and
Research, vol. 6, no. 3, pp. 196–199, 2021.
[2] S. Jadhav and R. Khanai, “Authentication And Counterfeit Detection
Of Currency Using Image Processing,” in International Conference
on Recent Innovations in Engineering and Management (ICRIEM-16),
2016, pp. 709–715.
[3] S. Shinde, L. Wadhwa, and D. Bhalke, “Feedforward back propaga-
tion neural network (FFBPNN) based approach for the identification
of handwritten math equations,” in Advances in Intelligent Systems
and Computing, Cham: Springer International Publishing, 2021, pp.
757–775.
[4] D. Kumar and S. Chauhan, “Indian fake Currency Detection using
computer vision,” International Research Journal of Engineering and
Technology, vol. 7, no. 5, pp. 2870–2874, 2020.
[5] S. U. Kadam, S. B. Shinde, Y. B. Gurav, S. B. Dambhare, and C.
R. Shewale, “A novel prediction model for compiler optimization with
hybrid meta-heuristic optimization algorithm,” Int. J. Adv. Comput. Sci.
Appl., vol. 13, no. 10, 2022.
[6] S. A. Amirsab, M. Mudassir, and M. Ismail, An automated recognition of
fake or destroyed Indian currency notes. International journal of advance
scientific research and engineering trends. 2017.
[7] Q. Zhang and W. Q. Yan, “Currency detection and recognition based
on deep learning,” in 2018 15th IEEE International Conference on
Advanced Video and Signal Based Surveillance (AVSS), 2018.

6
Authorized licensed use limited to: Marathwada Mitra Mandal's College of Engg. Downloaded on August 22,2024 at 08:47:57 UTC from IEEE Xplore. Restrictions apply.