Can computer vision and machine learning finally bring an end to the scourge of counterfeit products?
Few if any technology “solutions” to the problem have worked well up to now. Manufacturers and brands have deployed barcodes, QR codes, holographic stickers, serial numbers, etchings and other attempts at unique identification of legitimate products. But counterfeits continue to flood the market.
Last year, it was a $2.3 trillion problem worldwide, according to Roei Ganzarski, chief executive officer of Alitheon. Fake products ranged from the familiar Gucci bags and Rolex watches to pharmaceuticals, brakepads, seat belt assemblies and aircraft parts. Perhaps scariest of all, the U.S. Department of Energy earlier this year found counterfeit parts in U.S. nuclear plants.
Ganzarski likens the problem to a virus that keeps finding ways to circumvent the vaccine. Sometimes it doesn’t take much. Among the least effective methods of counterfeit prevention, he says, are tags or stickers that are affixed to the product. All counterfeiters have to do is fake the label of “authenticity.” In such cases, “you’re so confident that the product is real, you won’t even pay attention to what’s inside.”
Clearly, new methods are needed to separate authentic products from fakes. Ganzarski believes the solution lies in something that, at first blush, might seem suspiciously low-tech: pictures.
Can something that simple really authenticate every object in the world? It can, says Ganzarski, if one thinks of a photo as a digital fingerprint — every bit as unique as a literal human fingerprint.
He’s talking about more than taking a snapshot, even though a good smartphone camera of at least 12 megapixels — in short, just about any iPhone in service today — will suffice. What makes the everyday selfie powerful proof of authenticity is the application of algorithms and machine learning to identify the tiniest differences between real and counterfeit items.
Undetectable to the naked eye, “there are minute nuances and differences between physical items,” Ganzarski says. To minimize them, the engineering world has had to come up with machining quality standards such as Six Sigma. But counterfeiters aren’t able to keep within those strict tolerances, and the differences give them away.
Alitheon creates a “print” of an item and stores it digitally in the cloud, separate from the image itself. That way, a buyer can check on the validity of a given item several years later, by taking a picture and sending it in for comparison. The app then verifies whether, say, a microchip came off the legitimate manufacturer’s assembly line on a specific date, or whether the item in question is unfamiliar, and therefore in all likelihood fake.
One application of the technology that’s especially promising is the authentication of gold bars. Comparing a questionable bar with one that came from the proper manufacturer, Alitheon can detect the slightest variance in ridges, captured with the aid of an algorithm, down to the sub-pixel level. In the process, it can help put a stop to the circulation of gold that’s mined with forced or child labor.
Fingerprint systems in use today typically identify between seven and 11 points of commonality, while Alitheon’s system can detect more than 500 points on the same part, Ganzarski says.
For the system to work, of course, an authentic item must have been registered originally. But once it is, the app has established a baseline against which future iterations of the product can be tested.
One might see the process of photographing products as numbingly time-consuming, but Ganzarski insists that it’s scalable. It’s worth the effort, he says, given the huge amount of money that brands stand to lose when copycats flood the market.
That said, the technology is young. Alitheon has only been around for four years, and its product has been on the market for less than a year. “We spent a lot time making sure the technology actually works,” Ganzarski says. “You don’t want false positives.”
Initial applications are in consumer products, including high-value items and collectibles such as car parts, handbags and luxury watches. Other potential uses in the near future, Ganzarski says, are computer chips, especially those designed for military applications; high-value manufacturing, and automotive parts that are safety-related.
If it proves out at scale, the technology could indeed have a huge impact on safety. Detecting fake material in an airplane seat would be inconvenient; finding it in the wheel brakes is another matter entirely.
“What are the consequences of getting it wrong? You start with that,” says Ganzarski. “Eventually, though, why wouldn’t you want to know about every single part?”