On the design of a watermarking system: considerations and rationales Jean-Paul Linnartz, Geert Depovere, and Ton Kalker Philips Research Laboratories Prof. Holstlaan 4, WY8, 5656 AA Eindhoven, The Netherlands (linnartz, kalker,depovere)@natlab.research.philips.com Abstract. This paper summarizes considerations and rationales for the design of a watermark detector. In particular, we relate watermark detection to the problem of signal detection in the presence of (structured) noise. The paper builds on the mathematical results from several previously published papers (by our own research group or by others) to verify and support our discussion. In an attempt to unify the theoretical analysis of watermarking schemes, we propose several extensions which clarify the interrelations between the various schemes. New results include the matched filter with whitening, where we consider the effect of the image and watermark spectra and imperfect setting of filter coef- ficients. The paper reflects our practical experience in developing watermarking systems for DVD copy protection and broadcast monitoring. The aim of this pa- per is to further develop the insight in the performance of watermark detectors, to discuss appropriate models for their analysis and to provide an intuitive rationale for making design choices for a watermark detector. 1 Introduction The understanding of reliable methods to detect embedded data or watermarks has pro- gressed substantially over the past years. Many of the first proposals for watermarking emerged from the image processing community. It was also recognized that detection theory and spread-spectrum communication have several aspects in common with wa- termark detection, and results from these fields are now also exploited to improve the detection performance. This paper reviews the relation with detection theory and de- velops an intuitive understanding of the behavior of various approaches to watermark detection. It is not intended as a how to recipe, but rather as an attempt towards the development of a better understanding and a unified and more rigid theoretical model- ing of watermark detection. Hitherto, several detection principles have been proposed and verified experimentally, but theoretical support often was meagre. Our discussion mostly refers to theoretical models, rather than to experiments. Nonetheless most of the models have been verified by experiments reported in previous publications. New results are obtained to further verify and illustrate detection performance. Our exper- iments have been conducted during the development of watermarking systems, both for a consumer electronics application, viz., the JAWS system [?][?] proposed by the Millennium Group as a solution to DVD copy protection[?], and for a professional ap- plication, viz., the VIVA system for automated monitoring of advertisements and news