This thesis takes place in the field of Physical Object Security in view of the development of secure and reliable authentication schemes and is a part of the broader project of Information-theoretic analysis of deep identification systems.

Among the great variety of technologies available for the fight against object counterfeiting, a recent anticopy technology emerged in 2004 which showed great promise as an interesting trade-off offering high level of security against product counterfeiting while still maintaining a relative simplicity in implementation for mass-manufacturing industries. This technology is called the Copy Detection Pattern (CDP). The recent breakthrough in synthetic image generation based on deep learning architectures were shown to be a major threat to the classical methods used for the authentication of CDP. As such, there is a pressing need in developing open researches on the security guarantees provided by the CDP authentication scheme that would allow for reproducible experiments and fair comparison of its capabilities by independent researchers.

The main objective of this thesis is to provide an in-depth study of the printing-imaging channel associated to the CDP authentication framework both theoretically and experimentally so as to develop a better understanding of this technology’s main challenges. In this regard, this thesis proposes novel directions for the description and analysis of the printing-imaging channel of CDP, providing new tools and measures of the channel’s features and statistics. These new approaches are studied both from a theoretical point of view, developing sound foundations to discuss the problem of CDP-based framework security guarantees, and experimentally, by testing them on publicly available datasets of CDP. The thesis also adresses the question of applying these new tools to enhance the classical authentication scheme and demonstrates that the proposed novel approaches are actually able to detect even the strongest attacks on CDP known as of today.