| Steganography, or information hiding, is the science of concealing messages within innocuous-seeming cover data such as images, video or audio files. Watermarking refers to the use of steganography to exert control over the cover data itself, for example embedding copyright labels in a copyrighted image, or embedding signals to trace a copy of a song that is distributed over file-sharing networks. In either case the goal is preserving the hidden information while preventing it from being discovered. An adversary attacks the steganographic algorithm by distorting the cover data so as to erase any hidden messages or render them undetectable. The adversary also attempts to detect the hidden message data. In applications such as covert communication, any evidence whatsoever that steganographic communication is taking place constitutes a failure of the information hiding system. In watermarking applications the need for concealment is less strict, but any information about a watermark's structure can be used by the adversary to refine an attack.; This dissertation presents research results on several problems in steganography and watermarking. First, we present a solution to the problem of key exchange, using public-key cryptography, over a steganographic channel. Our solution includes a new method for creating a supraliminal channel, a side channel critical for inconspicuous key exchange. Along these lines we present solutions to the problem of zero-knowledge, or asymmetric watermark detection. This is the problem of proving that a watermark signal has been embedded in cover-data, but without providing information that would facilitate its removal. We implement these algorithms and develop attacks on asymmetric watermarks, as well as techniques for circumventing these attacks. Beyond these protocol-level results, we provide research results in steganalysis, the problem of detecting evidence of hidden messages, and reverse-engineering their structure. After steganalysis of an echo-hiding watermark proposed for the Secure Digital Music Initiative (SDMI), we developed improved detector structures for echo-hiding watermarks, and then general techniques for reverse-engineering watermarks through histogram analysis. |
