Physical goods counterfeiting affects over 7% of global trade, making it a major issue. While there have been a number of overt technologies, such as holograms and specialized barcodes, as well as covert technologies, such as taggants and PUFs, these solutions have had a limited impact on the market for counterfeit goods because of a variety of reasons, including cost, adoption barriers, and clonability. Entrupy present a novel method for differentiating between authentic and fake products by using machine learning algorithms to tiny photographs of real physical things.

Our system’s fundamental tenet is based on the notion that microscopic traits in a genuine product or a class of products (matching to the same larger product line) reveal intrinsic commonalities that can be exploited to separate real products from their corresponding counterfeit version. A wide-angle microscopy gadget that takes a microscopic image of a significant portion of a physical object is a crucial component of our system. We demonstrate that a highly accurate classification engine may be created using machine learning methods (ConvNets and bag of words) based on the tiny images that were recorded.

Based on the captured microscopic images, we demonstrate that one can create a highly accurate classification engine for differentiating the authentic versions of a product from the counterfeit ones using machine learning algorithms (ConvNets and bag of words). This property also holds for “super-fake” counterfeits observed in the market that are difficult to detect with the naked eye. We outline the architecture of an end-to-end physical authentication system that makes use of portable hardware, mobile devices, and a cloud-based ecosystem for object verification. Utilizing a sizable dataset of 3 million photos of diverse products and materials, including fabrics, leather, medicines, electronics, toys, and shoes, we test our method. To ensure the validity of common objects, the categorization accuracy is around 98%.