Cryptographic methods in multimedia identification and authentication

• Authors: Dan Boneh;Michael Todd Malkin
• Venue: Cryptographic methods in multimedia identification and authentication
• Year: 2006

This dissertation addresses several multimedia identification and authentication problems with concepts inspired by cryptography. Digital watermarking systems are used to authenticate ownership of and embed data in multimedia signals. Generally, it is assumed that the watermark embedding system and the watermark detection system share the same secret watermarking key and operate in a trusted environment. In real-world applications, however, this is often not the case. Watermark embedders operate in a trusted environment, but watermark detectors, such as DVD players, are usually run in a hostile environment. It is necessary, therefore, to ensure that the secret watermarking key does not become compromised in the hostile environment of the watermark detector. Our Secure Watermarking system solves this problem with the help of a secure module, for example a smartcard. We also consider the problem of robust image identification. Generally, images are assumed to be corrupted by natural processes such as noise. More sophisticated methods consider that an attacker may be behind these processes, usually assuming that the same attacks are involved but are calibrated more intelligently. We inject cryptographic concepts into this problem with the randlet transform. We randomly choose basis functions so an attacker can not take advantage the basis functions in mounting attacks, as it could, for example, with the wavelet transform. The randlet transform proves to be very successful at identifying images which have been attacked, and is especially successful against rotation and cropping attacks, two attacks which have generally been very difficult to overcome. Next, we apply the randlet transform to image watermarking, using a variation of the QIM watermarking system with non-orthogonal basis functions which are resistant to attacks. Again, the randlet transform makes it more difficult for an adversary to attack the system. Finally, we consider plagiarism detection. The problem of finding duplicate documents in a large database of documents is well-studied, but very little attention has been paid to intelligent plagiarists. We show that previous systems can easily be circumvented by certain attacks, and present a system that is robust to such attacks. Again, our solution relies on cryptographic concepts to achieve its success.