For binary alphabets, $n$ users and a false accusation probability of $\eta$, a code length of $m\approx \pi^2 c_0^2\ln(n/\eta)$ is provably sufficient to withstand collusion attacks of at most $c_0$ colluders. This improves Tardos' construction by a factor of $10$. Furthermore, invoking the Central Limit Theorem we show that even a code length of $m\approx \half\pi^2 c_0^2\ln(n/\eta)$ is sufficient in most cases. For $q$-ary alphabets, assuming the restricted digit model, the code size can be further reduced. Numerical results show that a reduction of 35\% is achievable for $q=3$ and 80\% for~$q=10$.
Category / Keywords: collusion-resistant watermarking Publication Info: Modified version has been submitted to Designs, Codes and Cryptography Date: received 9 Feb 2007 Contact author: boris skoric at philips com Available formats: Postscript (PS) | Compressed Postscript (PS.GZ) | PDF | BibTeX Citation Version: 20070214:104016 (All versions of this report) Discussion forum: Show discussion | Start new discussion