Efficient Watermark Detection and Collusion Security

Abstract

Watermarking techniques allow the tracing of pirated copies of data by modifying each copy as it is distributed, embedding hidden information into the data which identifies the owner of that copy. The owner of the original data can then identify the source of a pirated copy by reading out the hidden information present in that copy. Naturally, one would like these schemes to be as efficient as possible. Previous analyses easured efficiency in terms of the amount of data needed to allow any different copies to be distributed; in order to hide enough data to distinguish many users, the total original data must be sufficiently large. ere, we consider a different notion of efficiency: What resources does the watermark detector need in order to perform this tracing?

We address this question in two ways. First, we present a modified version of the CKLS media watermarking algorithm which improves the detector running time from linear to polylogarithmic in the number of users while still maintaining collusion-security. Second, we show that any public, invertible watermarking scheme secure against c colluding adversaries must have at least Ω(c) bits of secret information.