Improving the security of watermark public detectors

In this paper, we analyze the watermark detection when the decoder is publicly available. This situation puts new challenges of current watermark design to assure the decoder security, The common feature among most watermarking algorithms is that the decision boundary can be completely specified by a finite set of parameters. Hence, it can be estimated using a finite number of samples. In this paper, we describe a generalized attack for removing the watermark with minor perceptual distortion and establish its effectiveness using a direct implementation with least mean square (LMS) algorithm. The attack is abstract and can be applied to image, audio, and video watermarking schemes. Next, we describe a counterattack by using fractal decision boundaries, which are obtained by processing the original boundary. The resulting boundary cannot be parameterized and hence it cannot be estimated using a finite number of samples. The performance of the new technique is essentially similar to any watermarking algorithm from which it is derived.