Fooling Public-Key Watermarking Detectors with Optimal Color Noise

Public-key watermarking methods should have two properties: (1) allowing everyone to determine whether a watermark exists in an image or not; (2) having a high detection probability in case of malicious modification. This paper proposes a pollution attack which pollutes the watermark embedded in an image with an optimal color noise and hence fools the detector of the public-key watermarking scheme. Furthermore, it applies the proposed attack to public-key subspace watermarking schemes. In our experiments, pirated images are generated by inserting optimal colored signals into watermarked images. As a result, the pirated images are of high quality, but of low detection probability. The experiment results demonstrate that the present pollution attack is very effective.