Informed detection revisited

Abstract

Watermarking systems can employ either informed detection, where the original cover work is required, or blind detection, where it is not required. While early systems used informed detection, recent work has focused on blind detection, because it is considered more challenging and general. Further, recent work on "dirty-paper watermarking" has suggested that informed detection provides no benefits over blind detection. This paper discusses the dirty-paper assumptions and questions whether they apply to real-world watermarking. We discuss three basic ways in which an informed video-watermark detector, developed at Sarnoff, uses the original work: canceling interference between the cover work and the watermark, canceling subsequent distortions, and tailoring the watermark to the perceptual characteristics of the source. Of these, only the first is addressed by theoretical work on dirty-paper watermarking. Whether the other two can be accomplished equally well with blind watermarking is an open question. © Springer-Verlag Berlin Heidelberg 2005.