Collusion attacks for digital fingerprinting of image and video

Abstract

Copyright protection has become a great concern in the electronic distribution of digital multimedia. Digital watermarking is a technology proposed to help address this concern. To facilitate tracing of copyright violators, personalised watermarks or fingerprints containing user and/or transaction information can be embedded into the media data. However, significant threats to fingerprinting application come from a class of attacks known as the collusion attacks. Such attacks analyse a number of uniquely-fingerprinted data and creates a new copy for which fingerprint detection fails or performs poorly. In this thesis, three existing linear and non-linear collusion attacks on image/video fingerprinting, namely averaging collusion attack, maximum-minimum collusion attack, and Stone's zero-correlation collusion attack, are studied and analysed. Based on the study, two new collusion attacks, namely single-target zero-correlation collusion attack and multi-target zero-correlation collusion attack, are proposed. Mathematical models of their effectiveness in defeating fingerprint detection are also derived.