Estimation of Information Hiding Algorithms and Parameters

摘要

The goal of this project is the development of a system of useful tools for reverse-engineering covert channels and information hiding systems. This includes new algorithms for detection and estimation of certain hiding systems and the statistical artifacts they leave behind. One of our main observations is that severe false alarms leak a great deal of information about a watermark detector algorithm. The tendency to admit certain extreme false alarms a property we call superrobustness is an exploitable weakness in a detector. Using the techniques developed in this project, we participated in and won an international contest to defeat an unknown watermarking system. We did this by reverse-engineering the algorithm through the yes/no output of the watermark detector. Likewise, the participation in the contest spurred new research, in particular the 'noise caliper' technique of plumbing a detection region by growing false positives.