Public key watermarking by eigenvectors of linear transforms

Digital watermarks are signals embedded in multimedia data to allow copyright enforcement. In most watermarking schemes the embedded signal must be known for watermark detection, which leads to severe security risks. Van Schyndel et al. proposed a public watermark detection principle that works without explicit reference to the embedded signal. In this paper, extensions of this scheme are considered, and the applicability in practice is investigated. The new approaches are significantly less complex than the previously proposed scheme. Further, they are more robust against attacks via exhaustive search for the embedded watermark and against attacks that intend to confuse the public watermark detector. However, one drawback of all discussed schemes is the large signal length that is necessary for robust detection.