Counterfeit coins ‘detected more easily’ using image-mining

Counterfeit coins remain a threat to global currencies, with malicious actors flooding markets with fakes, but there are hopes imaging technology may make them easier to spot.

While no counterfeit is completely detection-proof – no matter how genuine it appears – there are always some tell-tale signals of forgery that may not be noticeable to the naked eye but can be picked up with close scrutiny.

In a new paper in the journal Expert Systems With Applications, researchers at the Centre for Pattern Recognition and Machine Intelligence (CENPARMI) at Concordia University present a novel framework that uses image-mining techniques and machine learning algorithms to identify flaws in counterfeit coins.

“Using image technology, we scanned both genuine and counterfeit coins so we can look for anomalies that may be either two- or three-dimensional, such as letters or the face of the person on the coin,” says Ching Suen, a professor in the Department of Computer Science and Software Engineering and the paper’s supervising author.

“This framework is not only about safeguarding our economy and resources,” adds the paper’s lead author, CENPARMI postdoctoral fellow Maryam Sharifi Rad (pictured above). “It is also about pushing the boundaries of technology and improving security.”

The researchers’ framework is built around fuzzy association rules mining. This approach uses artificial intelligence to find patterns that are similar but “fuzzy,” ie not clear enough to be exact copies. However, the framework will eventually arrive at a certain range of results where positive matches be confidently identified.

The method begins by using state-of-the-art scanners to examine coins suspected of being counterfeit. The coins are provided by law enforcement agencies.

The scanned images are then segmented to regions of interest, which consist of collections of localised coherent regions referred to as “blobs.” These blobs are recognized based on visual similarity and composition, which provide relevant features the researchers can extract. Blobs are like clues that help the researchers figure out what is going on in the scanned images.

Fuzzy association rules mining is performed using these blobs to extract frequent patterns from the images. These patterns capture relationships among the blobs’ attributes, such as colour, texture, shape and size. The patterns help researchers to better understand the images and tell whether a coin is real or fake.

The blobs play a crucial role in generating fuzzy association rules, which represent implicit knowledge and relationships within a set of images, aiding in image classification tasks.

The researchers say they believe their technique can be applied to detect all manner of counterfeit items beyond coins.

“This method can be used to detect all kinds of fake goods, which we are seeing all over the world,” says Suen. “It can also be used to detect fake labels on fruits, wines, liquor and so on. There are many places where this can be applied.”

The journal article can be viewed here.