A Similarity Attack to Correlation-Based Public Watermarking Detection

In this paper, we consider a security issue of ambiguity attack during public correlation-based watermark detection in a generic spread spectrum watermarking scheme, in which the probability distribution function (pdf) of the embedded signals can often be estimated. By comparing the changing behavior of the matching length among random sequences generated with different seeds from the same pdf, with that among signals extracted from the watermarked images, we find out that, 1) attacks like adding noises or JPEG compressions act as a shifting function on the pdf, resulting in distinct random sequences with a similar pdf; 2) the necessary conditions for this attack can be established between the matching length of random sequences and the extracted signals. Thus, the prover can cheat practically his ownership with another similar watermark and random number. Experimental results have shown the validity of the findings.