Professional Documents
Culture Documents
A Review On Currency Classification and Image To Text Conversion Methodologies
A Review On Currency Classification and Image To Text Conversion Methodologies
Abstract:- Currency classification and Image to Text measuring similarity, and ultimately recognizing the
OCR are essential technologies that find applications in character accurately.
various domains, including finance, retail, and
automation. The approach outlined in this paper has the Keywords:- Currency Recognition, CNN, OCR, Deep
potential to detect currencies from multiple countries. Learning.
However, for practical implementation purposes, the
focus is solely on Indian paper currencies. This system I. INTRODUCTION
offers the advantage of convenient currency checking at
any time and location, leveraging Convolutional Neural The World Health Organization (WHO) reports that
Networks (CNN) for effective implementation. Extensive globally, an estimated 285 million people are visually
testing was conducted on each denomination of Indian impaired, with the majority residing in developing nations.
currency, resulting in an impressive 95% accuracy rate. Among them, approximately 45 million individuals are
To further refine accuracy, a classification model was blind. Despite available solutions, none fully replicate the
developed, incorporating all pertinent factors discussed reading experience of sighted individuals, highlighting the
in the paper. Notably, the unique features of paper need for an affordable, portable text reader for the visually
currency play a pivotal role in the recognition process. impaired.
By emphasizing these elements and harnessing CNN
technology, the proposed system demonstrates significant A proposed solution involves creating a smart device
promise in accurately detecting and validating Indian with a multimodal system capable of converting any
paper currencies. It stands poised to serve various document into an accessible format. This device would
applications effectively. On the other hand, Image to enable blind individuals to read through tactile feedback and
Text OCR focuses on extracting text from images, auditory output via a text-to-speech engine, providing an
enabling the conversion of non- editable documents into experience akin to sighted individuals.
searchable and editable formats.
Visually impaired individuals encounter challenges in
Both technologies contribute to automation and daily tasks, including identifying items or information,
efficiency in handling diverse visual information. Optical leading to difficulties in navigating new environments. This
Character Recognition (OCR) is a technologydesigned to is especially problematic in scenarios like correctly
recognize and interpret both printed and handwritten identifying medications. Thus, innovative solutions such as
characters by scanning text images. This process involves smart devices and mobile applications are essential to
segmenting the text image into regions, isolating enhance accessibility and quality of life. Technology,
individual lines, and identifying each character along particularly mobile phones, plays a pivotal role in
with its spacing. After isolating individual characters facilitating communication and access to information for the
from the text image, the system conducts an analysis of visually impaired. Technologies like text-to-speech
their texture and topological attributes. This involves conversion and optical character recognition (OCR) enable
examining corner points, unique characteristics of effective interaction with computers through vocal
various regions within the characters, and calculating interfaces.
the ratio of character area to convex area Prior to
initiating recognition, the system creates templates that In conclusion, addressing challenges faced by visually
store the distinctive features of uppercase and lowercase impaired individuals in accessing information and
letters, digits, and symbols. navigating their environment is crucial. Innovative
technological solutions can significantly improve
These templates serve as reference models for accessibility and inclusion for this community.
comparison during the recognition phase. During
recognition, the system matches the extracted
character's texture and topological Features with those
stored in the templates to determine the exact character.
This matching process involves comparing features of
the extracted character with templates of all characters,
II. BACKGROUND AND RELATED WORK denominations exhibit distinct patterns when viewed with
infrared vision due to varying amounts of infrared light in
The idea we're presenting isn't entirely original, different areas. To accommodate the size differences of
especially regarding the recognition of Indian currency. Indian rupees, sensors are strategically placed at the
Previous attempts have been made to tackle this challenge. beginning and end lengths of the note for accurate detection.
However, our approach seeks to streamline the currency
recognition process, making it more efficient and less
resource- intensive. We aim to develop a solution that can
operate effectively on lower-end computing devices. In this
section, we will review past efforts in this field, identifying
their shortcomings and outlining areas where improvements
can be made.
Traditional Method
The Scanning Module's Layout: Fig 1:- The layout for sensor placement is illustrated
The primary function of the sensing unit is to collect
data from the input bill and forward it to a processing In the scanning module, different colors signify various
module. There are six emitter-sensor pairs in all. Where denominations, and the circle inside each color denotes the
each infrared (IR) emitter is matched with a photosensor. placement of the sensor. The size of Indian currency notes
These pairs are integrated within the device, positioned are specified in Table1 and the scanning module's sensors
opposite each other on both sides of a bill that was inserted. comply with these requirements. The start of each note is
To provide accurate and consistentreadings, the emitters face indicated by the black barrier on the left. The dimensions of
upward and the photosensors are positioned downward a 1000 rupee note are represented by the outside black
toward the banknote. boundary, while the dimensions of 500, 100, 50, 20, and 10
rupee notes are represented by the colors yellow, green,
To provide accurate and consistent readings, the pink, red, and orange, respectively.
emitters face upward and the photosensors are positioned
downward toward the banknote. Different Indian rupee
Pros: Provides tactile differentiation for various The next step for the system is to determine how
denominations, Simple and intuitive for blind individuals similar each feature is to the matching feature template
to use, doesn’t require additional equipment or linked to a certain denomination after it has been retrieved
technology, universally applicable to all blind from the input currency image. Higher similarity traits
individuals. receive a vote of one, whereas lower similarity features
Cons: Limited to physical currency only, doesn’t receive a vote of zero. The system then counts the number of
provide additional information like condition or features that got a single vote. If the count of features
authenticity. receiving a vote of one exceeds a certain threshold, the
currency is classified as known; otherwise, it is classified as
A. Feature Extraction unknown. The features extracted from the currency note are
Six unique characteristics are taken from each currency represented in Figure 2, while the architecture of the
note in the manner outlined for identifying paper money in proposed method is illustrated in Figure 3. This approach
Indian rupees. Two of these attributes are especially used to aims to effectively classify Indian rupee currency notes by
determine the currency's denomination, which helps the leveraging specific features and similarity calculations
system choose the right currency template.
algorithms. These machines can identify and sort different in documents and photos. Tesseract supports multiple
denominations based on various features, such as size, color, languagesand can be customized for specific
and security features. applications.
ATMs (Automated Teller Machines): ATMs are Adobe Acrobat OCR: Adobe Acrobat includes OCR
equipped with currency recognition capabilities to functionality for converting scanned documents into
authenticate and handle banknotes of different editable and searchable text. It is commonly used for
denominations. These systems employ a combination of document management anddigitization.
sensors and image processing techniques.
III. PROPOSED SYSTEM
Retail Automation Systems: Some retail environments
use automated systems that can recognize and process Develop and train an advanced machine learning model
various currencies during transactions, providing for currency classification. Utilize deep learning techniques,
efficiency and accuracy in cash handling. such as convolutional neural networks (CNNs), to improve
the system's ability to recognize various currencies,
Google Cloud Vision OCR: This OCR feature allows including different denominations and security features.
you to extract text from documents and images on Implement state-of-the-art OCR algorithms, including those
Google Cloud. It supports multiple languages, and the based on deep learning architectures, to improve the
extracted text can be used for various applications, such accuracy of text extraction from images. Consider
as document analysis and content indexing. techniques like attention mechanisms for handling complex
document layouts.
Tesseract OCR: Google created Tesseract, an open-
source OCR engine. It is frequently used to identify text System Architecture Currency Classification:
Convolutional Neural Networks Certainly! Here’s OCR Algorithm: Optical Character Recognition (OCR)
Convolutional Neural Net- works (CNNs) encode images is a computer vision process used for detecting and
into vector representations: interpreting text within images. It plays a vital role in
enabling Natural Language Processing algorithms to
Input Image: CNNs take an image as input. comprehend the content of documents.
Convolutional Layers: They use filters to extract features Input Layer: This layer consists of grayscale images
like edges and shapes. Activation Function: Apply representing text-containing documents. Output Layer:
functions to capture complex relationships. The identified text or characters are indicated by
binary or multi-class labels produced by the output
Pooling Layers: Down-sample to retain key information. layer.
Flattening: Convert feature maps into a1D vector.
Secret Layers: A fully connected neural network,
Fully Connected Layers: Capture relationships and pooling layers, ReLU (rectified linear unit) layers, and
output a vector. convolutional layers are some examples of these layers.
Output: The final vector represents the image and is used It's crucial to remember that Artificial Neural Networks
for tasks like classification and object detection. (ANNs), which are made up of many neurons, cannot
directly extract features from pictures. Convolutional and
pooling layers are used to address this. Though these layers [7]. Ms. Seema A. Dongare , Prof. Dhananjay B.
are quite good at extracting features, they are not appropriate Kshirsagar, Ms. Snehal V. Waghchaure, Handwritten
for jobs involving categorization. For categorization, a fully Devanagari Character Recognition using Neural Net-
linked neural network is therefore required. Understanding work, IOSR Journal of Computer Engineering (IOSR-
each section independently is crucial before delving further JCE) e-ISSN: 2278- 0661, p- ISSN: 2278-8727 Vol-
into these ideas. ume 16, Issue2, Ver. X (Mar-Apr.2014), PP 74-79 .
[8]. Mitrakshi B. Patil, Vaibhav Narawade, Recognition of
IV. CONCLUSION Handwritten Devna- gari Characters through
Segmentation and Arti cial neural networks, Interna-
In summary, the investigation of the categorization of tional Journal of Engineering Research and Technology
cash demonstrates their significant influence on a number of (IJERT) Vol. 1 Issue 6, August- 2012. ISSN: 2278-
facets. These innovations in machine learning and image 0181.
processing are revolutionizing the way we engage with [9]. Mandeep Kaur, Sanjeev Kumar, A RECOGNITION
visual data, streamlining workflows, and improving SYSTEM FOR HAND- WRITTEN GURMUKHI
accessibility. By automating the identification and CHARACTERS, International Journal of Engineer- ing
management of various currencies, currency classification Research and Technology (IJERT) Vol. 1 Issue 6,
simplifies cross-border trade, retail operations, and financial August - 2012 ISSN: 2278-0181.
transactions technology increases productivity by [10]. Miroslav NOHAJ, Rudolf JAKA, Image preprocessing
streamlining document management. Reducing manual labor for optical character recognition using neural networks,
and mistakes is achieved by data extraction and the Journal of Patter Recognition Research, 2011.
transformation of images into editable and searchable text. [11]. Nisha Sharma , Recognition for handwritten English
Systems for classifying currencies are essential to letters: A Re- view, In- ternational Journal of
international trade because they facilitate smooth Engineering and Innovative Technology (IJEIT)
transactions and eliminate obstacles brought on by currency Volume 2, Issue 7, January 2013.
variability. By enabling people with visual impairments to [12]. J.Pradeep, Diagonal based feature extraction for
access printed and handwritten information, OCR improves handwritten alphabets recog- nition System using
financial inclusion and fosters an inclusive digital neural network, International Journal of Computer Sci-
environment. ence and Infor- mation Technology (IJCSIT), Vol 3,
No 1, Feb 2011.
REFERENCES