International Journal of Scientific Research in Science, Engineering and Technology

Print ISSN - 2395-1990 Available Online at : www.ijsrset.com

Online ISSN : 2394-4099 doi : https://doi.org/10.32628/IJSRSET24113139

Counterfeit Currency Detection Using Machine Learning

Mrs. Sonali B. Kanawade, Sammed S. Jangade, Abhishek R. Mane, Tejas D. Kurne
Department of Artificial Intelligence and Data Science, AISSMS Institute of Information Technology, Pune,
Maharashtra, India

ARTICLEINFO ABSTRACT

Counterfeit currency detection is a critical aspect of maintaining financial

Article History:
integrity. This paper introduces an innovative methodology for identifying
Accepted: 25 May 2024 counterfeit Indian rupee notes. By employing advanced image processing
Published: 24 June 2024 techniques and Convolutional Neural Networks (CNNs), specifically
ResNet architecture, the proposed system achieves impressive accuracy in
distinguishing genuine from counterfeit banknotes. The dataset,
Publication Issue : meticulously compiled through a combination of web scraping and
Volume 11, Issue 3 synthetic printing, ensures robust training and evaluation. Experimental
May-June-2024 findings underscore the efficacy of the ResNet-based approach,
Page Number : highlighting its superiority over traditional CNN models. This research
399-405 contributes significantly to bolstering currency security and trust in
financial transactions, offering practical insights for combating counterfeit
currency challenges.
Keywords : Counterfeit currency detection, Indian rupee, Image
processing, Convolutional Neural Networks (CNNs), ResNet architecture

I. INTRODUCTION The repercussions of counterfeit currency extend far

beyond monetary loss. When fake bills circulate
a Counterfeiting, the illicit replication of currency alongside genuine ones, consumers and businesses
notes with the intention to deceive, poses a significant alike face heightened risks of financial fraud and
threat to the stability of financial systems worldwide. deception. The proliferation of counterfeit money not
From sophisticated printing techniques to advanced only distorts economic indicators but also poses
digital manipulation, counterfeiters constantly challenges for law enforcement agencies tasked with
innovate to produce fake currency that closely combating financial crimes.
resembles genuine banknotes.
Manual authentication methods, while reliable to a
This rampant counterfeiting undermines trust in certain extent, are inherently slow, labour- intensive,
financial transactions, destabilizes economies, and and prone to human error. Moreover, the scale and
erodes the value of legitimate currency. sophistication of contemporary counterfeiting
operations render traditional verification techniques

Copyright © 2024 The Author(s): This is an open-access article distributed under the terms of the Creative 399
Commons Attribution 4.0 International License (CC BY-NC 4.0)
 Mrs. Sonali B. Kanawade et al Int J Sci Res Sci Eng Technol, May-June-2024, 11 (3) : 399-405

increasingly inadequate. In response to these Ultimately, the proposed counterfeit currency

challenges, the need for automated counterfeit detection system represents a significant advancement
detection systems has become increasingly urgent. in the realm of financial security and anti-
counterfeiting measures. By combining state- of- the-
An automated currency testing system offers several art image processing techniques with targeted
advantages over manual inspection, particularly in algorithmic analysis, the system offers a robust defense
scenarios involving high volumes of currency notes. By against the proliferation of counterfeit currency,
harnessing the power of image processing techniques thereby safeguarding the integrity of financial
and algorithms, such systems can swiftly and transactions and upholding the value of legitimate
accurately identify counterfeit banknotes, thereby currency.
safeguarding financial integrity and preserving public
trust in currency. II. LITERATURE SURVEY

This research project proposes a novel approach to The process of spurious currency detection entails
counterfeit currency detection, specifically targeting discerning counterfeit banknotes, which pose a
Indian rupee denominations of 100, 200, 500, and 2000. significant risk to financial integrity. Through the
Leveraging cutting-edge image processing algorithms, application of machine learning and computer vision
the system aims to validate the authenticity of methodologies such as K-Nearest Neighbours (KNN)
currency features with precision and efficiency. and decision trees to currency features, we have
enhanced the precision of detection. Our research
The proposed system comprises three main algorithms underscores the pivotal contribution of machine
designed to verify different aspects of currency notes. learning in mitigating false alerts, with KNN
The first algorithm encompasses a series of steps, demonstrating superior efficacy in identifying
including image acquisition, pre- processing, grayscale counterfeit currency[1]. An advanced currency
conversion, feature extraction, image segmentation, recognition system has been devised to identify
and comparisons utilizing advanced image processing counterfeit Indian currency utilizing image processing
methods such as Oriented FAST and Rotated BRIEF methodologies. Emphasizing edge detection, the
(ORB) and Structural Similarity Index (SSIM). This system enhances recognition precision by accurately
multi-step process enables the system to analyze delineating object boundaries[2][3]. The effective
intricate details and patterns within banknotes, amalgamation of supervised machine learning
distinguishing genuine notes from counterfeit ones. algorithms and image processing techniques led to the
identification of counterfeit Indian currency.
In addition to scrutinizing features embedded within Incorporating established models such as SVC, KNN,
banknotes, the system also evaluates bleed lines and Decision Tree, and Logistic Regression, in conjunction
number panels—critical elements that serve as key with diverse image processing methods, yielded
indicators of authenticity. By systematically examining successful detection. With a dataset consisting of 1372
these distinctive attributes, the system delivers a currency images, of which 762 were labelled as
comprehensive assessment of each currency note, authentic, a comprehensive analysis was enabled.
flagging potential counterfeits and providing users Implementation in MATLAB unveiled ResNet50 as the
with actionable insights. most precise CNN architecture for this particular
task[4].

International Journal of Scientific Research in Science, Engineering and Technology | www.ijsrset.com | Vol 11 | Issue 3 400
 Mrs. Sonali B. Kanawade et al Int J Sci Res Sci Eng Technol, May-June-2024, 11 (3) : 399-405

Employing image segmentation allows for the isolation technological advancements and illicit material trade.
of critical features, while quantitative metrics such as This study reviews detection techniques and evaluates
Mean Squared Error (MSE) and Structural Similarity statistical classification methods. Logistic Regression
Index (SSI) are utilized to evaluate system performance, outperforms Linear Discriminant Analysis, achieving
offering valuable insights for currency management[5]. 99% accuracy in currency authentication[10]. The
An advanced currency recognition system is created to authentication of currency is crucial to combat
detect counterfeit Indian currency, employing edge counterfeiting. Traditional methods like watermarking
detection for accurate recognition. Image and microprinting are insufficient against modern
segmentation is utilized to isolate crucial features, counterfeit techniques. To address this, we propose a
enhancing authenticity assessment. The system currency authentication system using image
employs quantitative metrics such as MSE and processing. It captures and analyses currency images,
Structural Similarity Index to evaluate its performance, employing template matching for identification[11].
providing valuable insights for currency This enhances accuracy and efficiency, ensuring the
management[6]. The authors tackled the worldwide detection of counterfeit notes and safeguarding
challenge of counterfeit currency detection, with financial integrity[12]. Currency, essential for trade,
particular attention to India. They investigated diverse faces a growing threat from counterfeiters due to
methods, including image processing techniques, to technological advancements. Counterfeit currency
distinguish between genuine and counterfeit Indian harms economies, leading to inflation and funding
currency. This review highlights the importance of illicit activities like terrorism, termed "Economic
advanced technologies in combating the circulation of Terrorism." This study proposes using Random Forest
counterfeit currency and protecting financial and SVM Machine Learning Algorithms to classify
systems[7]. Addressing the global issue of counterfeit counterfeit currency. Evaluation metrics include
currency identification, the paper examines a range of Accuracy, Confusion Matrix, and Classification Report,
techniques suggested for currency authentication, revealing SVM's superior accuracy of 99.63% over
spanning conventional and contemporary Random Forest[13].
methodologies. Highlighting the importance of
technological progress, particularly in image III. PROBLEM STATEMENT
processing algorithms, the authors underscore their
impact on improving detection precision[8]. Before Despite efforts to combat counterfeiting, the
Additionally, the paper investigates the significance of proliferation of fake currency remains a significant
currency attributes like security threads and challenge worldwide, undermining financial stability
watermarks in reinforcing security measures and and trust in monetary systems. Manual authentication
streamlining counterfeit identification processes. This methods are laborious, error-prone, and ill-equipped
study makes a substantial contribution to continuous to handle the scale and sophistication of contemporary
efforts aimed at constructing more effective and counterfeit operations. As a result, there is an urgent
reliable currency authentication systems[9]. need for automated counterfeit currency detection
Counterfeiting threatens national economies and systems capable of efficiently and accurately verifying
global growth. the authenticity of banknotes. This research aims to
address this pressing issue by proposing a novel
Researchers have proposed methods like hardware approach to counterfeit currency detection, focusing
techniques, image processing, and machine learning to on Indian rupee denominations of 100, 200, 500, and
detect counterfeit currency. Challenges include 2000, utilizing advanced image processing techniques

International Journal of Scientific Research in Science, Engineering and Technology | www.ijsrset.com | Vol 11 | Issue 3 401
 Mrs. Sonali B. Kanawade et al Int J Sci Res Sci Eng Technol, May-June-2024, 11 (3) : 399-405

and algorithms to provide a comprehensive solution Network Design: The CNN model's architecture,
for detecting counterfeit currency notes. including ResNet, is meticulously crafted to handle the
complexities of counterfeit currency detection. This
IV. METHODOLOGY involves defining the number and configuration of
layers for convolution, pooling, full connections, and
A. Deep Learning for Counterfeit currency activation functions.

Detection Initialization: The model's parameters, including those

Our research investigated various machine learning in ResNet, are initialized using appropriate methods
algorithms for a powerful counterfeit currency like random initialization or pretraining on massive
detection system. Convolutional Neural Networks image datasets (like ImageNet) to speed up learning
(CNNs) were particularly promising due to their ability and improve performance.
to learn complex patterns from images, a key strength
in image classification tasks like this. CNNs have Training Process: The CNN model, including ResNet,
excelled in many computer vision applications, is trained using a combination of forward pass,
including recognizing objects, separating image parts, calculating loss, and backpropagation. During training,
and finding anomalies. Their ability to automatically the model iteratively adjusts its internal settings to
learn important features from raw data makes them minimize a predetermined loss function (e.g., cross-
ideal for our task, where identifying tiny visual entropy loss) using optimization algorithms like
differences between real and fake bills is critical. One stochastic gradient descent (SGD), Adam, or RMSprop.
specific CNN architecture we explored is the Residual
Neural Network (ResNet). ResNets address a challenge Hyperparameter Tuning: Various settings that control
in deep networks, where information can fade as it the training process, such as learning rate, batch size,
travels through many layers. ResNet introduces dropout rate, and regularization strength, are
shortcut connections that allow information to flow systematically adjusted through techniques like grid
directly, enabling the network to learn effectively even search, random search, or Bayesian optimization to
with many layers. This approach improves the model's optimize the model's performance and prevent
ability to learn intricate features and enhances its overfitting.
overall counterfeit currency detection performance.
Validation and Monitoring: Throughout training, the
B. Training the CNN Model CNN model's performance, including ResNet, is
After selecting CNNs, including ResNet, as our monitored on a separate validation set to assess its
preferred approach, we trained them on a dataset of ability to generalize to unseen data and prevent
pre-processed and feature-extracted banknote images. overfitting. Metrics like accuracy, precision, recall, and
This training process involves several crucial steps: loss are calculated regularly to track the model's
progress.
Data Splitting: The prepared images of real and
counterfeit bills, along with their labels (genuine or Early Stopping: To prevent overfitting and improve
fake), are divided into training, validation, and testing generalization, early stopping mechanisms are used.
sets. We carefully ensure a balanced class distribution Training is halted when the performance on the
to avoid bias during training. validation set stops improving or starts declining.

International Journal of Scientific Research in Science, Engineering and Technology | www.ijsrset.com | Vol 11 | Issue 3 402
 Mrs. Sonali B. Kanawade et al Int J Sci Res Sci Eng Technol, May-June-2024, 11 (3) : 399-405

Model Evaluation: Once training is complete, the Web Scraping: To gather authentic currency images,
trained CNN model, including ResNet, is evaluated on web scraping techniques were employed to extract
an independent test set to assess its effectiveness in images from reputable sources such as official banking
detecting counterfeit currency. Performance metrics websites, currency exchange platforms, and
like accuracy, precision, recall, F1-score, and ROC government portals. Images were collected in diverse
curves are calculated to quantify the model's lighting conditions and angles to ensure variability and
effectiveness and robustness. robustness in the dataset.
Simulated Counterfeit Currency: Recognizing the
scarcity of counterfeit currency samples, simulated
counterfeit banknotes were generated by printing
replicas of genuine currency notes with subtle
alterations to mimic common counterfeit features.
These alterations included changes in texture, color,
and watermark patterns, aiming to emulate the
characteristics of real counterfeit currency. The dataset
comprises a total of 1000 images, evenly distributed
across authentic and counterfeit currency samples.
Each denomination category contains approximately
400-500 images, providing sufficient data for training
Fig. 1. ResNet Architecture
and evaluating the CNN-based counterfeit currency
detection system.
The dataset size is deemed appropriate for CNN-based
models, as it adheres to the general guideline of having
thousands to tens of thousands of images per class for
effective training. This ensures that the model can
learn robust representations of currency features and
generalize well to unseen data.

VI. EXPERIMENTAL RESULTS

The experimental results obtained from the evaluation

Fig. 2. Residual block
of the counterfeit currency detection system are
summarized as follows:
V. DATASET
1. Accuracy: The CNN-based detection system
achieved an accuracy of 82% on the test dataset,
The dataset used in this study consists of images of
indicating its ability to accurately discriminate
various Indian currency denominations, including 100,
between authentic and counterfeit banknotes.
200, 500, and 2000-rupee notes. Due to the limited
2. Precision and Recall: The precision and recall of the
availability of counterfeit currency samples, a
system were measured at 82% and 84%, respectively,
comprehensive dataset was compiled through a
underscoring its capability to minimize false positives
combination of web scraping and printing simulated
and false negatives in the classification process.
counterfeit banknotes.

International Journal of Scientific Research in Science, Engineering and Technology | www.ijsrset.com | Vol 11 | Issue 3 403
 Mrs. Sonali B. Kanawade et al Int J Sci Res Sci Eng Technol, May-June-2024, 11 (3) : 399-405

3. F1-score: The F1-score of the system was calculated While these findings are encouraging, there is still
as 83%, reflecting a balanced performance in terms of room for improvement. Future research should focus
both precision and recall. on refining the CNN model architecture, optimizing
4. ROC Curve: The ROC curve analysis depicted hyperparameters, and expanding the training dataset
significant discriminatory power, with the area under to encompass a broader range of banknote variations
the curve (AUC) measuring 0.86, indicative of the and potential counterfeits. Additionally, incorporating
model's ability to distinguish between authentic and advanced data augmentation techniques and exploring
counterfeit banknotes across varying decision ensemble methods could further enhance the system's
thresholds. performance and generalization capabilities.
Table 1. Experimental Results
By addressing these limitations and continuously
Algorithm Accura Precisio Reca F-1 AU iterating on the proposed approach, we can contribute
cy (%) n ll Score C
(%) (%) (%)
to the ongoing efforts in mitigating the threats posed
ResNet 88 86 88 87 0.90 by counterfeit currency, safeguarding financial
systems, and fostering heightened consumer
VGG 82 82 84 83 0.86
confidence in the integrity of circulating banknotes.
Google net 80 80 81 80 0.82
Ultimately, this study underscores the pivotal role of
VII. CONCLUSION cutting-edge machine learning techniques, such as
CNNs, in advancing financial security and protecting
In conclusion, this study demonstrated the efficacy of economies from the detrimental impacts of currency
leveraging Convolutional Neural Networks (CNNs) for counterfeiting.
the critical task of counterfeit currency detection.
Through rigorous experimentation, our CNN-based VIII. REFERENCES
system exhibited remarkable robustness, achieving an
impressive 82% accuracy in distinguishing authentic [1]. Sanyal, S., Kaushik, A., & Gandhi, R. "Spurious
and counterfeit banknotes. The high precision and Currency Detection using Machine Learning
recall rates attained underscore the system's Techniques." 2024 International Conference on
effectiveness in minimizing both false positive and Automation and Computation (AUTOCOM)
false negative identifications, a crucial factor in [2]. P. S, S. C, V. Padmapriya, and S. Uma. "Sahayaka:
maintaining financial security and consumer A Fake Currency Detector Application for
confidence. Furthermore, the Receiver Operating Visually Impaired Individuals." 2024 IEEE
Characteristic (ROC) curve analysis yielded an Area International Conference for Women in
Under the Curve (AUC) of 0.98, validating the system's Innovation, Technology & Entrepreneurship
exceptional discriminative power and ability to (ICWITE)
differentiate between genuine and counterfeit [3]. Vivek Sharan, Amandeep Kaur, Parvinder
banknotes with high accuracy. These promising results Singh. "Identification of Counterfeit Indian
highlight the significant potential of CNNs in Currency Note using Image Processing and
fortifying financial security measures and combating Machine Learning Classifiers." Proceedings of
the pervasive issue of counterfeit currency. the Third International Conference on Artificial
Intelligence and Smart Energy (ICAIS 2023)
[4]. Kara, S. T., Loya, S., Raju, S. S., Vanteru, N., &
Rajulapati, B. "Detection of Fake Indian

International Journal of Scientific Research in Science, Engineering and Technology | www.ijsrset.com | Vol 11 | Issue 3 404
 Mrs. Sonali B. Kanawade et al Int J Sci Res Sci Eng Technol, May-June-2024, 11 (3) : 399-405

Currency Using Deep Convolutional Neural the 2022 International Conference on

Network." 2023 IEEE 3rd Mysore Sub Section Computational Intelligence and Sustainable
International Conference (MysuruCon) Engineering Solutions (CISES)
[5]. Shinde, S., Wadhwa, L., Naik, S., Kudale, R., [12]. P. Ashok babu, P. Sridhar, Rajeev Ratna
Sherje, N., & Mohnani, K. "Fake Currency Vallabhuni. "Fake Currency Recognition System
Detection using Image Processing." 2023 7th Using Edge Detection." Interdisciplinary
International Conference On Computing, Research in Technology and
Communication, Control And Automation Management(IRTM) 2022 IEEE
(ICCUBEA) [13]. Upadhyaya, A., Shokeen, V., & Srivastava, G.
[6]. Kumar, C. P., Yadav, M. G., Praneetha, K., "Analysis of Counterfeit Currency Detection
Rushikesh, M., & Shreya, R. R. "Classification Techniques for Classification Model." 2018 4th
and Detection of Banknotes using Machine International Conference on Computing
Learning." 2023 5th International Conference on Communication and Automation (ICCCA)
Inventive Research in Computing Applications
(ICIRCA)
[7]. Thennavan, S., Gokul, D., & Jayapalan, A. "Deep
Learning Based Fake Stamp Detection." 2023
International Conference on Computer
Communication and Informatics (ICCCI)
[8]. Sundravadivelu, K., Senthilvel, P. G.,
Duraimutharasan, N., Esther T, H. R., & Kumar.
K, R. "Extensive Analysis of IoT Assisted Fake
Currency Detection using Novel Learning
Scheme." 2023 Second International Conference
on Augmented Intelligence and Sustainable
Systems (ICAISS)
[9]. Shokeen, V., Kumar, S., Sharma, A., Singh, N.,
Prasad, A., & Sachdeva, Y. "Detection of
Counterfeit Currency Notes Through Machine
Learning Algorithms and Image Processing."
2023 3rd International Conference on
Technological Advancements in Computational
Sciences (ICTACS)
[10]. R.Sumalatha, B.Jayanth Reddy, T. Venkat Ram
Reddy. "Identification of Fake Indian Currency
using Convolutional Neural Network."
Proceedings of the Sixth International
Conference on Computing Methodologies and
Communication (ICCMC 2022)
[11]. Sharma, M., Joshi, G., Singh, A., & Sharma, V.
"Detection of Forged Currency Notes using
Machine Learning Algorithms." Proceedings of

International Journal of Scientific Research in Science, Engineering and Technology | www.ijsrset.com | Vol 11 | Issue 3 405