Error- and collusion-secure fingerprinting for digital data

Fingerprinting means making copies of the same data identifiable by hiding additional information (a fingerprint) in the data. Embedding the additional data can be done by watermarking techniques, which are mainly a branch of signal processing. Most watermarking methods, however, do not treat colluding adversaries who have obtained more than one copy, compare their copies, see differences and use this information to make a copy without a fingerprint. Therefore, there are cryptographic fingerprinting methods to obtain collusion tolerance on top of a watermarking layer. But the most important fingerprinting method by Boneh and Shaw excludes a priori any errors on the watermarking layer, i.e., it is assumed that no changes to the fingerprint occur except those based on the information from collusion attacks. This is a stronger assumption than what most underlying watermarking schemes offer. This assumption can be justified by making each individual mark fault-tolerant on the watermarking layer, e.g., by replication, but that would imply a significant increase in the data size needed. Instead, here we implement the fault tolerance more efficiently on the cryptographic layer by generalizing Boneh and Shaw's fingerprinting methods. Our remaining assumption on the underlying watermarking is quite reasonable for watermarking methods that would be built according to the best currently known principles.