Abstract

In our daily lives, we use coins to pay for goods and services. However, the market for antique and historical coins is another place where coin quality and genuinity are important. Since counterfeiting has become more common as a result of technological advancements, dealing with fake coins is unavoidable. As a result, researchers have considered various methods in coin detection studies. In recent years, image-based coin detection has made extensive use of 2-D and 3-D image processing approaches. We propose a method for detecting counterfeit coins based on image content in this paper. We used
SIFT, SURF, and MSER to determine the degree of similarity between our datasets. Then, using statistical analysis, we determine which descriptor is the most effective criterion for
counterfeit coin detection. SIFT was chosen as the most reliable algorithm for the Danish and Canadian coin image dataset according to the Experiments' results. The autoencoder is
then trained to detect anomalies in the coin images. A coin image is fed to the trained autoencoder as input and outputs a new image. Using the chosen criterion, the output image is compared to a baseline image. If the similarity between these two images is greater than a certain threshold, the coin is genuine. For training, most counterfeit coin detection methods require fake data. Our autoencoding-based anomaly method can eliminate this. Our
proposed method for distinguishing genuine coins from counterfeit coins yielded promising results. In addition, we present a method for increasing the speed of counterfeit coin detection. We conducted our research on the Mint Mark of Canadian toonies coin images and we were able to achieve acceptable results by combining the edge detection technique with GAN and autoencoder.