Random Codes for Digital Fingerprinting

Unlicensed, or illegal, copying of data is a problem in many areas. Despite various efforts in copy protection and copyright enforcement (both legal and technical), the problem still exists, and with the growing use of digital means of storing and distributing data the problem seems to be getting worse. Storing a unique, invisible marking in each distributed object is a possible way of dealing with the problem of illegal copying. That way, if an illegal copy is found somewhere, it is possible to find the original holder of the object and to take legal action. This type of copyright enforcement scheme is called fingerprinting. This thesis deals with the use of random codes for fingerprinting of digital data in the context of several users colluding to create an untraceable, illegal copy (such users are called pirates). Some theoretical bounds on the performance of such codes are presented. A general method to statistically test whether certain users have taken part in the creation of a particular illegal copy is presented, and pirate strategies that prevent this kind of testing are derived. Finally an explicit testing method is presented that tests whether a certain, proposed group of users should be considered guilty of having created a specific illegal fingerprint. The method's performance is analyzed, and it is found that the error probability is practical for an interesting range of values of the number of colluding pirates and the length of the fingerprints. ii Acknowledgments