Statistically Secure Anti-Collusion Code Design for Median Attack Robustness for Practical Fingerprinting

Abstract

Digital fingerprinting is a technique to prevent customers from redistributing multimedia contents illegally. Main attack for fingerprinting is the collusion attack, where multiple users collude by creating an average or median of their individual fingerprinted copies, and escape identification. Previous research such as ACC (anti-collusion code) cannot support large number of users, and also vulnerable to LCCA (linear combination collusion attack). We present a practical SACC (scalable ACC) scheme to generate codebooks for supporting large number of users; and angular decoding scheme to be robust on LCCA. We implemented the SACC codebook using a Gaussian distributed random variable for various attack robustness, and the fingerprint embedding using human visual system based watermarking scheme. We experimented with standard test images for collusion detection performance, and it shows good collusion detection performance over average, median attacks. For LCCA collusion attack on SACC, our angular decoding scheme identifies the correct colluder set under various WNR (watermark to noise ratio).