Machine Learning Algorithms for Identifying Fake Currencies

Abstract

Currency in the form of coins and notes is extensively used in numerous aspects of our day-to-day lives, including but not limited to parking meters, telephone booths, and so on. People collect coins and notes for a variety of reasons, including the fact that they are used as currency and have financial value, as well as the reality that coins and notes often have artistic value and may give a vivid glimpse into the community interaction that existed in the past. Despite this, there has been an increase in the number over the course of the last several years of illicit counterfeiting networks that produce and sell false coins while at the same time printing bogus note currency. These activities have resulted in significant financial loss and have harmed society as a whole. Therefore, the ability to recognise counterfeit currency is really necessary. This work presents an innovative method for identifying counterfeit Indian notes based on their images. To identify the fake currency, a novel research approach is introduced to meet the existing gap using machine learning algorithms. A vector space constitutes the space for dissimilarities that is produced by doing a comparison between an image and a group of prototypes; this space is where a currency image is represented. Each dimension determines how unlike an image is to a given prototype by comparing it to the image that is being considered. To determine the degree to which two currency pictures are dissimilar to one another, the local key spots on each image must first be located and characterized. If the characteristics of the notes are used as a basis, it will be possible to find the matching key areas between the two images in a time-efficient manner. As an additional measure, a post-processing approach is suggested for the removal of mismatched critical points. One-class learning is used to detect false cash, because there are so few counterfeit bills in circulation; hence, only real bills are required to train the classifier. This is because there are so few counterfeit bills in circulation. In this work, a novel feature-based intensity calculation and classification approach is introduced. The machine learning thresholding, K-means, and support vector machine (SVM) algorithms were used. These algorithms were used to find the fake currency based on intensity values. The proposed method was tested on 50 images and obtained an accuracy of 96%, which showed that the classification of currency using machine learning algorithms provided better results compared to the existing approaches.