A Pollution Attack to Public-key Watermarking Schemes

Public-key watermarking schemes are required to possess two desirable properties: allowing everyone to determine whether a watermark exists in an image or not and ensuring high detection probability in case of malicious modification. In this paper we propose an attack which pollutes the watermark embedded in an image with an optimal colored noise so as to fool the detector of the underlying public-key watermarking scheme. We further show how to apply the proposed pollution attack to public-key subspace watermarking schemes to generate pirated images of high quality but of low detection probability. Our experiment results demonstrate that the proposed pollution attack is very effective.