Abstract
Fake bills cause a serious problem to the economy. In the past, color printing was not possible; however, recent advances in printing technology have allowed printing bills that are very similar to the original bills. This leads to a higher chance of having fake bills in circulation. ATMs use image comparison techniques to detect the currency value and the state of the bill, but this is not enough to check if the bill is real or fake. Thus, ATMs should have a framework that can distinguish fake bills in less time and more effectively. In this project, we propose an efficient image processing technique that will extract various features from the U.S. currency. The features that can be extracted using image processing algorithms are dimensions, color, written letters, security thread, texture, federal reserve indicator, serial number, and various other details. These are several indicators that will help to increase the accuracy of detecting fake bills in ATMs. In this project, we use discrete wavelet transformation (DWT) for texture extraction, optical character recognition (OCR) for reading the serial number and federal reserve indicator, and cross-correlation for finding similarities between the real and the received bill. We experiment with techniques such as machine learning and deep neural networks to extract more features and improve accuracy. This application works in a faster way, without requiring any additional hardware except a camera.