Abstract

should not affect the quality of data. Robustness refers to Downlthe requirement that embedded watermarks should be de-tected reliably even after some attacks. In the digital watermarking world, geometric attacks are regarded as very strong and harmful attacks. In most wa-termarking schemes, including the spread spectrum watermarking,3,4 the synchronization of an embedded wa-termark with the reference watermark is crucial for the wa-termark detection. Geometric attacks prevent watermark detection by desynchronizing the embedded watermark. Even slight rotation or scaling of the marked image can cause watermark detection failure if the watermark detector does not have a synchronization mechanism. One watermarking method that handles geometric at-tacks is to embed a watermark into a geometric-invariant domain. The Fourier-Mellin-transform-based method5,6 be-longs to this approach. This method seems to provide a nice Autocorrelation function ~ACF!-based watermarking is yet another approach.12–14 In this scheme, a periodic water-mark pattern is embedded into an image. The periodic wa-termark makes periodic peaks in the ACF of the watermark. The geometric transforms that have been applied to the marked image are estimated by inspecting the pattern of the extracted peak. This scheme also has the two detection fail-ure modes. For reliable watermark detection, both the au-tocorrelation AC peaks and the embedded watermark should survive attacks. Watermarking systems are often required to resist com-bined geometric and removal attacks. The removal attacks refer to some processing that can remove or attenuate an embedded watermark signal. The printing-scanning pro-cess, which is one of the most popular geometric attacks, is also a kind of combined attack. During the printing-scanning process, slight rotation or scaling of the image isImproved autocorre watermarking wit