Very early work on watermarking was essentially heuristic, in part, because watermarking was not recognized as a distinct technology. This began to change in the late 1980’s and early 1990’s when a number of published papers described a variety of different watermarking algorithms. A more rigorous understanding of watermarking then began to be developed, beginning in the mid-1990’s.

Perhaps the most significant progress has been in the development of increasingly sophisticated models of watermarking. In the early 1990’s it became common to model watermarking as a communications channel in which the cover Work and any subsequent distortions between the time of embedding and detection were treated as noise. The constraint of imperceptibility was met by imposing a global power constraint at the embedder.

In these early systems the added watermark signal is independent of the cover Work and we refer to this as blind embedding. Similarly, blind detection refers to the detection of a watermark signal in a cover Work, the detection being independent of the unwatermarked Work.