Anti-counterfeiting technology through artificial intelligence and challenge-response pairs

Information security has witnessed a tremendous growth in the last decade, covering a wide range of applications in our daily life from health, economy to intellectual property. Accordingly, security systems with strong anti-counterfeiting capability are crucial for protecting personal information and privacy. Conventional cryptographic primitives are mostly designed based on mathematical one-way functions, which possess high risk of information extrusion. To overcome the weakness, optical physical unclonable functions (PUF) has been spotlighted as a promising tool in authentication and identification for its simple detection and rich optical features. Many types of photonic materials have thus been developed into PUF over the past decade by taking advantage of its complex optical signals. In this thesis, we used liquid crystal droplets and silver nanocubes to fabricate microscale and nanoscale optical PUF. Different algorithms for facile recognition and authentication were developed and studied. In the first part, we developed a deep learning network to achieve anti-counterfeiting based on multicolored liquid crystal droplets. In the second part, two-factor authentication are applied to fluorescence image generated from nanocubes, which can improve the accuracy of authentication and is an enhancement of PUFs decoding process. Finally, we compared and discuss the pros and cons of using binary methods and deep learning methods for anti-counterfeiting technologies.Master of Science (Electronics