AI Spotlight: Using Machine Learning to Detect Counterfeit Goods

Share |

Pam is walking around Shanghai and sees shops that contain shoes, handbags, clothes, and jewelry that are all emblazoned with the logos of famous designers.  Upon walking inside, she notices that many of these products, while not very cheap, are still markedly cheaper than if she were to buy the exact same product in the United States.  The quality looks and feels great, so without a second thought, she purchases a handbag.  After all, this isn’t a shady stand on a roadside.  This is a seemingly legitimate store.  It isn’t until she’s back in the United States that she discovers that the Louis Vuitton handbag that she bought is a fake.

This type of situation is fairly common.  There are often only subtle differences between a real and fake product that may not be detectable at first touch or first look.  Other times, people are aware that what they’re purchasing is a counterfeit good and simply don’t care.  Regardless of the reason, counterfeiting is a serious problem, not just in the United States, but worldwide.

According to a 2016 report by the Organization for Economic Cooperation and Development (OECD) and the EU Intellectual Property Office (EUIPO), global imports of counterfeit goods are worth about 461 billion dollars, amounting to roughly 2.5 percent of total global imports.  Of all countries whose intellectual property rights are infringed upon, twenty percent of the total value of seizures comes from the United States.  Additionally, 63 percent of these fake goods originate in China.  These numbers are truly staggering and yet only account for the total value of physical goods, excluding online piracy.  To tackle this problem, researchers at New York University have developed a device called Entrupy to detect if a physical good is real or counterfeit.

Entrupy is a device that can scan the surface of different physical goods to check if they are real or counterfeit.  It combines machine learning with microscopy with an accuracy rate of over 98 percent.  It utilizes a dataset of three million images of various objects.  According to Ashlesh Sharma, the co-founder of Entrupy, the algorithms “pick up data points from millions of microscopic images looking for qualities like texture, contrast, topology, geometric shapes, thread counts, minor manufacturing artifacts such as scratches in the hardware stamps, wear, and many more details that you wouldn’t be able to see.  These details are fed into our custom machine learning pipeline, allowing us to determine a product’s authenticity.”  This high accuracy rate is improved upon with each new scan.

The benefit of using Entrupy over a traditional method of counterfeit-detection is that traditional methods tend to be invasive and sometimes damage the original good.  Additionally, sometimes retailers and wholesalers are not certain that the items that they are vending are real or counterfeit.  In fact, the main use of Entrupy is for vendors to determine authenticity rather than an individual consumer.  It has been so successful that it has authenticated 14 million dollars worth of goods.  It has also received 2.6 million dollars in funding from a team of investors.

SIIA agrees with the view that machine learning as a supplement to human work has high economic potential and often times creates new opportunity instead of displacing human work.  Entrupy is yet another example of machine learning as a tool to supplement human work.  It does not contribute to job displacement.  In fact, it helps the people responsible for authenticating products do their jobs better.  

While 14 million dollars is but a dent in the 461 billion dollars in counterfeit imports, the machine learning technology used by Entrupy represents potential for solving this problem.  It presents opportunity to find criminals who are knowingly producing and vending counterfeit goods, and stop lawful vendors from unknowingly vending these goods.  It prevents consumers from purchasing counterfeit goods and saves companies from having their IP infringed.  There will always be demand for counterfeit items, but the more pervasive that this machine learning technology is with prevention, the easier it will be to find and stop the sale and purchase of counterfeit items.

Diane Diane Pinto is the Public Policy Coordinator at SIIA. Follow the Policy team on Twitter @SIIAPolicy.