Authentic Vision’s Machine Learning And Computer Vision Algorithms Enable Instant Identification Of Counterfeit Products Around The Globe
FOR IMMEDIATE RELEASE
How Authentic Vision utilizes computer vision and optical machine learning to identify counterfeit goods and deliver best-in-class user experience
Salzburg, Austria. July 9, 2020: Counterfeiting is big business. Globally, the Organisation for Economic Co-operation and Development (OECD) and EU’s Intellectual Property Office estimate that counterfeit goods account for more than $500 billion (USD), or more than 2.5%, of global trade. Counterfeiters are getting increasingly sophisticated, with many able to replicate the look and feel of genuine products. And while the performance, operating and safety characteristics of these fakes fail to meet the specifications of a genuine product, we can easily be fooled by these products – until they fail.
Authentic Vision, a global leader in product authentication, understands the challenges inherent in identifying genuine products and the risks fakes create. Using machine learning and other advanced technologies, they can validate authentic products from fakes instantly and keep us safe.
The process of determining if a specific product is genuine or fake is complex. Sophisticated counterfeiters can source similar raw materials, copy the design of complex products, reproduce packaging and logos, and even slip counterfeit products into distribution channels, making it extremely difficult to identify fake products. For some products, such as agrochemicals, industrial fluids and consumables, there is an even more stealthy threat – adulteration. Their original manufacturers design these products with specific formulations and compositions that can often be easily be modified, diluted or adulterated before arriving in the hands of buyers.
Purchasing a fake designer handbag or belt might offend your ego and empty your pocketbook. But, counterfeiters also fabricate fake brake components that endanger your safety, adulterated agrochemicals that cause irreparable damage to the environment, and industrial parts and fluids that cause significant physical and financial loss. Counterfeiters’ advances have made identifying fake goods and products extremely difficult. Antiquated guidelines, such as the visual inspection of products, closely examining trademarks, identifying printing and linguistic errors on products and packaging, looking for missing product accessories, and pricing discrepancies, are wholly ineffective against modern counterfeits.
Perhaps even more concerning is that many of the ‘advanced’ anti-counterfeiting solutions available today are also ineffective, easily replicated or hacked by sophisticated counterfeiters.
The reality is that identifying counterfeit products requires both expert-level knowledge of the products themselves and how they are designed, manufactured, and sold. The testing of counterfeits requires an expensive and sophisticated set of tools. But even if you had the appropriate expertise and required tools; the process to test and verify products as authentic would be lengthy. Some tests are destructive, so even with the knowledge, tools, training and time required to verify products as authentic, you would destroy the very products you wanted to purchase in the process of authenticating them.
The best solution for authenticating genuine products tags them at the manufacturer and can validate them as authentic instantly in real-time for the end-user. But, implementing this type of solution is complex.
Authentic Vision’s solution utilizes a patented tag applied to products or packaging. The patented label has several unique features that, when combined with machine learning and the Check-if-Real smartphone app, can authenticate products as genuine instantly.
No other authentication solution leverages these capabilities to quickly, easily and securely authenticate genuine products in real time with only a smartphone – a claim no other authentication solution can make. Let’s examine some of the technology that makes this work and where machine learning comes into play.
The Authentic Vision tag consists of two components: a patented random holographic “fingerprint” image and a digitally printed unique label identifier. The patented holographic ‘fingerprint’ image is a critical element in the solutions overall cryptographic security. The unique label identifier is based upon a serial number and cryptographic hash encapsulated in a backend database, making it unguessable, unique, irreproducible, yet verifiable.
Once the Authentic Vision label is attached to a genuine product, typically before it leaves the manufacturing facility, the challenge of identifying an authentic product morphs into validating the label itself as authentic. This is when machine learning plays a crucial role. Authentic Vision’s primary machine learning task is one of computer vision. Whenever the Authentic Vision tag is scanned with Authentic Vision’s CheckIfReal smartphone app, the app must validate the tag and inform the user whether they have an authentic or counterfeit product.
In this context, positive identification of either a genuine or fake product is critically important for the user experience – the best user experience instantly tells the user if they have a genuine or counterfeit product. Anything less than this positive identification, such as a response of “unable to confirm authenticity,” may leave end-users thinking “authentication doesn’t work, but it doesn’t say counterfeit either, so maybe we’re ok.” This result would be confusing to the end-user.
Visual machine learning algorithms learn to identify authentic and counterfeit tags by a process called training, presenting a neural net with tags identified as genuine. The neural net can learn the right underlying patterns and be able to identify a new, never before seen tag as genuine. Authentic Vision is able to make some assumptions about an authentic tags holographic image because it knows the specifications of the label and how a user might scan the holographic image.
Identifying a fake tag is more challenging. Some fake tags are nearly indistinguishable from authentic tags to the human eye, so correctly identifying fake tags used in training the neural net can be difficult. Further, the different ways counterfeiters try to duplicate the patented holographic ‘fingerprint’ image is infinite. Thus, it is impossible to know all the different versions of fake tags that might be found in the field. One approach utilized by Authentic Vision to identify fake tags is anomaly detection. The neural net learns very well how authentic tags appear and if a scanned tag appears dissimilar, it is declared as an anomaly and therefore counterfeit.
As the neural net is learning Authentic Vision’s patented labels, it might also unintentionally learn features of the camera and smartphone; aberrations in the camera hardware that are too minute for the human eye to discern or image processing optimizations typically used to improve the image for human visual perception rather than machine learning. Image processing algorithms embedded within smartphones can introduce digital noise or artifacts due to image compression. Other optimizations may attempt to make an image look more pleasing to our eyes.
Curating a tag dataset structured in a way that the neural net learns useful features only and does not learn undesired artifacts found in the tag dataset is an art. If the dataset is wrongly structured, the neural net may exhibit very good performance yet not work well in a real-world scenario with new tags. This capability, for the neural net to work reliably with samples not present in the training set is known as generalization. Authentic Vision excels is achieving good generalization to millions of labels with comparatively small datasets of only a few hundred labels. Dataset recording is both costly and arduous.
The machine learning algorithms developed by Authentic Vision must also be optimized to run on an incredibly wide range of smartphones, each with differing hardware, storage, and imaging capabilities. The algorithms are designed to operate within a minimal footprint yet provide maximum accuracy using an intelligent mix of machine learning and traditional, hand engineered computer vision algorithms.
Computer vision machine learning models often exceed 1Gb in size, but through the use of carefully designed classifiers and pipelines, Authentic Vision has been able to package these models to run on devices with small footprints.
The app that reads Authentic Vision’s tag is called CheckIfReal. The app runs on every smartphone – an incredible technological achievement in itself, considering it does heavy lifting computer vision on the device. (Check the list of smartphones here). This delivers on our founder’s vision to make real-time authentication possible for anyone with a simple scan from nearly anywhere in the world.
About Authentic Vision
Authentic Vision provides anti-counterfeiting and authentication technologies designed to protect your organization’s investments in product innovation, brand value and reputation while creating new opportunities to increase trust and engagement with consumers. The company’s unique holographic fingerprint tag, mobile authentication app and real-time analytics capabilities protect physical assets from counterfeiting and alert brand and product owners to potential fraudulent activity. Their anti-counterfeiting and authentication technologies help to minimize lost revenues and mitigate liability due to counterfeits and create new opportunities to engage with consumers through loyalty programs, incentives and future experiences that bridge the physical and digital. Visit AuthenticVision.com to learn more about their solutions, view case studies, download their consumer app or read the latest insights on how anti-counterfeiting and authentication technologies can help your organization.
Gernot Kalchgruber, Authentic Vision