UCL Discovery

Abstract

We analyze four recently proposed watermarking schemes for intellectual property protection of digital designs. The first scheme watermarks solutions of a hard optimization problem, namely the graph coloring problem. The other three schemes belong to a family of techniques for watermarking digital circuits on programmable hardware. They are different from the usual image and audio watermarking since they must maintain correctness of the watermarked objects. Thus their watermarks cannot be embedded in the form of small errors as usually done in audio and visual watermarking. Although constraint-based watermarking schemes existed long before, these schemes are the first ones to protect hardware designs. In this paper, we apply a novel method to break the first of these schemes. We show how to modify a watermarked object in such a way that every signature strings can be extracted from it. Thus anyone can claim ownership of the object, yet leave no traces of who leaked the object. According to our best knowledge, this method is new and it may be of its own interest. In the remaining three schemes, we show how to locate and to remove the,watermark embedded in the object, without knowing the secret key used in the embedding.