Is asymmetric watermarking necessary or sufficient?

In public watermarking applications, the public must have the ability to detect watermarks, but must not have the ability to remove them. It has been proposed that such systems might be made secure by using asymmetric key watermarking, in which the embedder and detector use different keys. This paper asks whether asymmetric-key watermarking is sufficient and necessary for secure public watermarking applications. The answer found, both to the question of sufficiency and the question of necessity, is basically no. Asymmetric-key watermarking, by itself, is demonstrably insufficient, because there exist asymmetrickey watermarking systems that do not provide the required security. Asymmetric-key watermarking might not be necessary, because it is possible to imagine irreversible watermarking systems, in which even complete knowledge of the embedder would not help an adversary remove watermarks. The paper concludes by suggesting that research in secure public watermarking should focus on the design of the detector. The embedding algorithm, and whether it employs any information that must be kept secret, is of secondary concern.