Watermark estimation through detector analysis

A watermark is a perceptually unobtrusive signal embedded in an image, an audio or video clip, or any other multimedia asset. Its purpose is to be a label which is holographically attached to the content. Moreover, it can only be removed by malicious and deliberate attacks (without a great loss of content quality) if some secret parameter K is known. In contrast, a watermark should be readily detectable by electronic means. This implies that electronic watermark detection is only feasible if the watermark detector is aware of the secret K. In many watermarking business scenarios the watermark detector will be available to the public as a block box D. The following question is therefore justified: can the secret K be deduced from the operation of the black box D? And if yes, what is the complexity of this process? In this paper we will address these questions for a large class of watermarking schemes. This work is an extension of earlier work at Philips Research [1].