On Collusion Attack for Digital Fingerprinting

Collusion attack is a challenge issue for digital fingerprinting, and it is very beneficial to study for fingerprinting design. Two models for linear and non-linear collusion attack are constructed from the angles of owner and colluder in the paper. Through theory and simulation analysis in fingerprints embedding domain and non-embedding domain some conclusions are obtained: (1) The attack effects in embedding domain and non-embedding domain are same for average collusion attack; (2) The attack effects in embedding domain is stronger than that in non-embedding domain for maximum attack, minimum attack and randneg attack; (3) The attack effects in embedding domain and non-embedding domain are similar for median attack, min-max attack and modneg attack; (4) The two models have the similar attacks effects in embedding domain. We have also studied the relationship between the quality of colluded image and collusion attack. Large experimental results prove the correctness of the received conclusions, which is very helpful for fingerprinting design and system analysis.