A New Collusion Attack and Its Performance Evaluation

Abstract

Digital watermarking is a technology proposed to help address the concern of copyright protection for digital content. To facilitate tracing of copyright violators, different watermarks carrying information about the transaction or content recipient can be embedded into multimedia content before distribution. Such form of “personalised” watermark is called “fingerprint”. A powerful attack against digital fingerprinting is the collusion attack, in which different fingerprinted copies of same host data are jointly processed to remove the fingerprints or hinder their detection. This paper first studies a number of existing collusion attack schemes against image fingerprinting. A new collusion attack scheme is then proposed and evaluated, both analytically and empirically. Attack performance in terms of fingerprint detectability and visual quality degradation after attack is assessed. Results obtained from spread spectrum fingerprinting experiments show that the proposed attack can impede fingerprint detection using as few as three fingerprinted images without introducing noticeable visual degradation, hence it is more powerful than those reported in literature. It is also found that increasing the fingerprint embedding strength and spreading factor do not help resist such malicious attacks.