The abundance of digital media and the availability of popular software tools to edit, reformat, and manipulate these media has raised fundamental issues of authenticity, trust, and forensic analysis. This project explores the mathematical foundations for these problems. Applications areas include information protection and law enforcement. The project is highly multidisciplinary and involves a synergy between research and educational activities in signal processing, communications, coding theory, information theory, and learning theory.
Specifically, the project develops an analytical framework based on fundamental statistical principles (Bayesian and learning theory) to develop novel algorithms and to characterize the reliability of the resulting authentication and forensic analyses. The project focuses on three research thrusts: 1) Desynchronization-Resilient Authentication. The ultimate limits of resilience to desynchronization are investigated, building on recent advances in unsupervised learning using graphical models. 2) Forensic Analysis. A framework is developed for Region of Interest (ROI) identification for images, based on Vapnik-Chervonenkis theory. 3) Blind Traitor-Tracing. Theory and codes for traitor tracing (aka digital fingerprinting) are developed for problems where the original signal to be protected is not available to the receiver.
