Volume 8, Issue 4, April – 2023 International Journal of Innovative Science and Research Technology

ISSN No:-2456-2165

Indian Currency Fake Note Detection System

Using Resnet 50
Riddhi Shinde1 Aishwarya Thorat2 Aditi Yadav3
Rajiv Gandhi Institute of Rajiv Gandhi Institute of Rajiv Gandhi Institute of
Technology, Mumbai Technology, Mumbai Technology, Mumbai

Jyoti Singh4 Rahul Jiwane5

Rajiv Gandhi Institute of Technology, Mumbai Department of Information Technology
Rajiv Gandhi Institute of Technology, Mumbai

Abstract:- The printing and scanning industry has made Overall, this project represents an important
significant technological advancements, but contribution to the field of counterfeit currency detection,
unfortunately, this has also led to a rise in counterfeiting. demonstrating the potential of machine learning to address
Counterfeit currency can negatively impact the economy this challenging problem. By improving the efficiency
and reduce the value of genuine money. Therefore, and accuracy of counterfeit currency detection, we hope to
detecting fake currency is crucial. Traditional methods contribute to the prevention of financial crime and the
have relied on hardware and image processing protection of the global economy.
techniques, which can be inefficient and time-
consuming. To address this issue, we have proposed a II. LITERATURE SURVERY
new approach that uses a deep convolutional neural
network to detect counterfeit currency. Our method The feasibility of a software-assisted counterfeit
analyzes currency images and can efficiently identify currency detection system is demonstrated through a
fake currency in real time. We trained a transfer learned detailed methodology for extracting existing features of
convolutional neural network using a dataset of two banknotes using image processing. The Bangladeshi bank
thousand currency notes to learn the feature map of notes are analyzed to identify the following features for
genuine currency. Once the feature map is learned, the testing purposes: micro- printing, watermark, optically
network is able to identify counterfeit currency quickly variable ink, iridescent ink, security thread, and ultraviolet
and accurately. Our proposed approach is highly lines, with the BDT 1000 parameters used in the analysis.
effective and significantly reduces the time required to To detect counterfeit currency, the proposed approach
identify fake currency among the 500 notes in our extracts multiple features from Indian currency through
dataset. various image processing algorithms, with theimage acquired
using an image acquisition device. Template matching is
I. INTRODUCTION then used to identify fake currency. To improve the accuracy
of the system, different parameters will be used to recognize
Counterfeit currency is a growing problem that affects the difference between fake and original currency notes
economies around the world. The use of advanced printing through image processing techniques.[1]
and scanning technology has made it easier for criminals to
produce fake currency, which can reduce the value of Convolutional Neural Networks (CNNs) are a type of
genuine money and cause significant economic damage. In neural network architecture commonly used in image
response to this issue, researchers have developed various processing tasks, such as image classification and object
methods to detect counterfeit currency, with machine detection. They use convolutional layers to automatically
learning being a particularlypromising approach. learn and extract features from the input image, which are
then used for classification or other tasks.
In this project, we propose a novel method for
detecting counterfeit currency using machine learning. The paper "Neural Networks for Image Analysis and
Specifically, we use a deep convolutional neural network to Processing" focuses on applying neural network algorithms
analyze images of currency notes and identify fake ones. to image processing tasks such as feature detection,
Our approach involves training the network using a large segmentation, and classification. The authors explore
dataset of genuine currency images, enabling it to learn the different neural network architectures and techniques for
features that distinguish genuine notes from fake ones. Once feature extraction, including the use of integral images for
trained, the network can quickly and accurately identify convolution, and simplify these methods to their essential
counterfeit currency in real-time, providing a valuable tool components. The result is a new approach to feature
for law enforcement agencies, financial institutions, and detection and matching, which can be applied in various
other organizations. areas of technology.[8]
To address the issue of counterfeit notes, various

IJISRT23APR1823 www.ijisrt.com 1604

Volume 8, Issue 4, April – 2023 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
minutiae parameters will be utilized, along with deep B. System Block Diagram
learning algorithms, to effectively distinguish between fake
and genuinecurrency notes. This will be achieved through the
application of image processing techniques such as
Convolutional Neural Networks (CNN), pattern recognition,
and RESNET50, amongother algorithms.

III. METHODOLOGY

Motivated by recent advancements in image processing

and the increasing availability of low-cost image acquisition
devices, we propose an approach for detecting counterfeit
currency using Convolutional Neural Networks (CNN). Our
approach involves extracting multiple features from Indian
currency and using them for fake currency detection. The
system architecture outlining our proposed work is
presented below. To identify fake currency, the proposed
approach acquires an image using an image acquisition Fig 2 System Block Diagram
technique and extracts security features from it using
ResNet50 a Convolutional Neural Networks algorithm. C. Work flow
Template matching is then performed to differentiate
between real and counterfeit currency. The approach is
unique in that it applies image processing to extract security
features from currency images and uses multiple security
features instead of just one.

A. Currency Features
The features that have been extracted thus far can be
classified as general features, which are features that are not
specific to any application, such as texture, color, contrast,
and shape. The specific features that are used for counterfeit
currency detection systems vary depending on the notes of
each country. In the case of Indian currency notes, the
following features are considered:

Fig 3 Steps of Implemented System

 The first step is to acquire the image of the currency

note using a digital camera and preserving it under UV
light.
 Pre-processing the image is done before data analysis
Fig 1 Security Features of Indian Currency and information extraction can take place. This
includes resizing the image.
Feature extraction is a process of obtaining relevant  The RGB image is converted to grayscale to extract
information or features from an image by applying image intensity information. Edge detection is applied to
processing algorithms. In the context of detecting fake identify the boundaries of the grayscale image.
currency, images of a currency note are acquired using a  Image segmentation involves dividing the image into
digital camera or scanner. Once the image is acquired, it multiple sections or crops.
undergoes pre-processing to enhance its quality, and then  Edge-based segmentation is used for feature extraction.
feature extraction is performed to extract specific features  The intensity of each extracted feature is computed, and
that can distinguish between genuine and counterfeit notes. if the intensity is above 70%, the note is considered
Both these steps are explained in detail in this section. genuine. Otherwise, it is considered a forgery.
 The conclusion is based on the intensity of all retrieved
characteristics.

IJISRT23APR1823 www.ijisrt.com 1605

Volume 8, Issue 4, April – 2023 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
 The Categorization Procedure is Outlined in the Stages  Load the Pre-Trained ResNet-50 Model: ResNet-50 is a
below. pre-trained deep CNN model that has been trained on a
large dataset for image classification. We can load this
 Pre-Processing pre-trained model into our code using popular deep
In pre-processing the images for fake currency learning libraries such as TensorFlow, Keras or
detection using ResNet-50, we typically perform the PyTorch.
following steps:
 Remove the Fully Connected Layer: The pre-trained
 Image Resizing: ResNet-50 is a deep CNN architecture ResNet-50 model typically ends with a fully connected
that was designed for image recognition tasks on large layer that outputs the predicted class probabilities for
datasets. Hence, it is important to resize the input the input image. However, for feature extraction, we
images to a fixed size that ResNet-50 can handle want to remove this layer and extract the output of the
efficiently. A typical size used for ResNet-50 is 224x224 last convolutional layer as features.
pixels.
 Forward Pass through the Network: Pass the pre-
 Normalization: To normalize the pixel values of the processed input image through the ResNet-50 model up
images, we can subtract the mean pixel value of the to the last convolutional layer. This will produce a
training set from each pixel and divide it by the tensor of feature maps that captures the image's high-
standard deviation. This helps to bring the pixel values level visual patterns.
of the images to a similar scale and can improve the
accuracy of the model.  Global Average Pooling: ResNet-50's final
convolutional layer produces a feature map of size
 Data Augmentation: Data augmentation is a useful 7x7x2048. To extract features, we perform global
technique to improve the performance of a machine average pooling over this feature map, which results ina
learning model. It involves creating new training data 2048-dimensional feature vector for each image.
by applying different transformations to the existing
data. Popular data augmentation techniques include  Feature Extraction: We can extract features for all
random cropping, flipping, rotation, and zooming, images in the dataset using the above steps and store
which can help increase the diversity of the training set them in a separate file or database. These features can
and make the model more robust to variations in the be used as inputs to a classifier model that is trained to
input data. distinguish between genuine and counterfeit currency
notes.
 Class Balancing: Ensure that the dataset used for
training ResNet-50 has a balance between the number Overall, feature extraction using ResNet-50 can be an
of genuine and counterfeit currency notes. A dataset effective way to extract discriminative features from
with imbalanced classes can result in a biased model images, even when the dataset is small. These features can
that performs poorly on the minority class. then be used to train a classifier that can accurately classify
genuine and counterfeit currency notes.
 Data Split: The dataset should be partitioned into three
subsets: training, validation, and testing. The ResNet- IV. CONCLUSION
50 model should be trained using the training set while
the validation set should be utilized to fine-tune the In conclusion, fake currency detection using machine
hyperparameters of the model. Finally, the testing set learning techniques like ResNet-50 is a promising area of
should be used to assess the model's performance. development that can help enhance the security of financial
transactions. By leveraging the power of deep learning
Overall, pre-processing is an important step in fake algorithms, it is possible to accurately detect counterfeit
currency detection using ResNet-50 as it can improve the currency notes and prevent their circulation. While ResNet-
accuracy of the model and help to avoid overfitting. 50 has been shown to be a highly effective algorithm for
fake currency detection, there are several other algorithms
 Feature Extraction that can be used for this task, and their performance can
Feature extraction is a crucial step in fake currency vary depending on the specific problem being solved.
detection using ResNet-50, where we extract relevant
features from the images and use them to train a classifier Advancements in technology, integration with
that can distinguish between genuine and counterfeit existing systems, multi-currency support, real-time
currency notes. Here are the steps involved in feature detection, mobile applications, and enhancing accessibility
extraction using ResNet-50: are some of the potential areas of development. As
technology continues to evolve, we can expect even more
accurate and efficient ways to detect counterfeit currency
notes, enhancing the security of financial transactions and
reducing the risk of counterfeit notes being circulated.

IJISRT23APR1823 www.ijisrt.com 1606

Volume 8, Issue 4, April – 2023 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
FUTURE SCOPE

The future scope of fake currency detection using

machine learning techniques like ResNet-50 is vast and
promising. Here are some potential areas of development:

The integration of fake currency detection algorithms

in mobile applications can be a game-changer, especially in
countries where counterfeit notes are widespread. This can
help users to easily and quickly check the authenticity of
currency notes using their mobile phones. Real-time
detection of fake currency notes using machine learning
algorithms can be useful in scenarios like ATMs and banks,
where large numbers of notes are processed every day.
Integration of fake currency detection systems with existing
systems such as ATMs, vending machines, and point-of-
sale terminals can enhance the security of transactions and
reduce the risk of counterfeit notes being circulated.
Overall, the future scope for fake currency detection
systems is promising, and there are several avenues for
further development and improvement. As technology
continues to evolve, we can expect even more accurate and
efficient ways to detect counterfeit currency notes.

REFERENCES

[1]. Zahid Ahmed, Sabina Yasmin, Md Nahidul Islam,

Raihan Uddin Ahmed “Image Processing Based
Feature Extraction of Bangladesh Banknotes.”
[2]. "Detection of Fake Indian Currency Using Machine
Learning Techniques." International Journal of
Research in Engineering and Technology, vol. 5, no.
8, 2016, pp. 302-307.
[3]. "Detection of Counterfeit Currency Using Machine
Learning Techniques." 2018 IEEE 2nd International
Conference on Inventive Systems and Control,
pp.152-157.
[4]. "Currency Recognition Using Machine Learning."
International Journal of Computer Applications, vol.
143, no. 6, 2016, pp. 22-28.
[5]. "Banknote Recognition Based on Convolutional
Neural Networks." 2019 International Conference on
Robotics, Control and Automation, 2019, pp. 92-96.
[6]. "Banknote Recognition Using Convolutional Neural
Networks and Principal Component Analysis." 2020
14th International Conference on Application of
Information and Communication Technologies,
2020, pp. 1-5.
[7]. "Fake Indian Currency Detection Using Machine
Learning Techniques." 2017 International
Conference on Inventive Communication and
Computational Technologies, 2017, pp. 580-584.
[8]. N. Rathee, A. Kadian, and R. Sachdeva, “Feature
fusion for fake Indian currency detection,”
Computing for Sustainable Global Development
(INDIACom), 2016 3rd International Conference on
IEEE, 2016.

IJISRT23APR1823 www.ijisrt.com 1607