Indian Currency Classification and Counterfeit Detection Using Deep Learning and Image Processing Approach

Abstract

These days, there is rampant usage of fake currency notes in every country, which is not good for the economy of that country. Further, the other side of the issue is the problem of visually impaired people to recognize bank notes. To address both of these problems, in this work we propose a deep learning model, which not only recognizes the real and fake Indian currency notes but also tells its correct denomination. We have employed various deep learning architectures for this task. In specific, we explored the applications of pre-trained deep learning models, like VGG16, GoogLeNet and MobileNet for currency classification and fake currency detection. The use of pre-trained deep learning models is motivated by the fact that these models are based on transfer learning and hence not data-hungry. As we had limited datasets available for this work, the pre-trained deep learning models were expected to give better performance. After evaluating the proposed approach on around 2572 images belonging to 6 denominations of Rs. 10, 20, 50, 100, 500, 2000, it was found that VGG16 obtained the highest classification as well fake detection accuracy of 98.08% and 97.95%, respectively. We further followed a conventional image processing-based approach and analyzed the discriminative feature differences between a real and a fake note. Various methods of preprocessing of the input image like edge detection, intensity mapping and HSV space conversion are done and the differences between the real and fake notes were analyzed, which can be used further for fake currency detection.