Synchronization-insensitive Video Watermarking using Structured Noise Pattern I. Setyawan, G. Kakes, R. L. Lagendijk Delft University of Technology, Faculty of Information Technology and Systems, Information and Communication Theory Group, Delft, The Netherlands {i.setyawan, g.kakes, r.l.lagendijk}@its.tudelft.nl ABSTRACT For most watermarking methods, preserving the synchronization between the watermark embedded in a digital data (image, audio or video) and the watermark detector is critical to the success of the watermark detection process. Many digital watermarking attacks exploit this fact by disturbing the synchronization of the watermark and the watermark detector, and thus disabling proper watermark detection without having to actually remove the watermark from the data. Some techniques have been proposed in the literature to deal with this problem. Most of these techniques employ methods to reverse the distortion caused by the attack and then try to detect the watermark from the repaired data. In this paper, we propose a watermarking technique that is not sensitive to synchronization. This technique uses a structured noise pattern and embeds the watermark payload into the geometrical structure of the embedded pattern. Keywords: Copyright protection, invisible video & image watermarking, geometrical transformation attack, watermark synchronization 1. INTRODUCTION One of the most difficult task faced by image and video digital watermarking algorithm developers is resistance to geometrical attacks. Geometrical transformations may take simple forms such as rotation, translation and scaling. They could also take much more complex forms, for example rubber-sheet stretching. A famous example of the latter is the StirMark software package that can perform a wide range of minor, unnoticeable geometrical transformation on an image including slight stretching, bending or shifting 1 . Another example of geometrical attack is the Digital Cinema Attack 2 . In this scenario, the geometrical attack is not actually performed on the watermarked data directly. Instead, the attacker records a (watermarked) movie being shown on the cinema screen using a camera. The result of this recording is then illegally distributed on Video CD’s or put on the Internet for download. The quality of the recording is influenced by numerous factors, e.g., the position of the camera with respect to the cinema screen, the (usually low) quality of the lenses in the camera and the fact that the cinema screen itself is not perfectly flat. All these factors contribute to the complex combination of geometrical transformation applied to the recorded video. Geometrical transformation attacks do not actually remove the watermark from the data. Instead, they work by exploiting the fact that most watermarking techniques rely on the synchronization between the watermark and the watermark detector. If this synchronization is destroyed, the detector can no longer correctly detect the presence of the watermark in the data although the watermark itself (or a major part thereof) might still remain in the data. As a simple example, let us take a simple image watermarking algorithm that embeds a pseudo-random noise pattern all over the image and detects the embedded watermark by correlating the watermarked image with an identical pseudo-random noise pattern 3 . A high correlation value will signify the presence of the watermark. If this watermarked image is now cropped by removing 5% of the pixels, the correlation value produced by the detector would be very low although most of embedded noise pattern itself is still present in the image. Some approaches have been developed in the literature to deal with attacks that destroy the synchronization between the watermark and the detector. In this paper, we present a new watermarking approach that is less sensitive to Copyright 2002 Society of Photo-Optical Instrumentation Engineers This paper will be published in the Proceedings of the SPIE, Security and Watermarking of Multimedia Contents IV, Vol. 4675, 2002 and is made available as an electronic preprint with permission of SPIE. One print or electronic copy may be made for personal use only. Systematic or multiple reproduction, distribution to multiple locations, via electronic or other means, duplication of any material in this paper for a fee or for commercial purposes or modification of the content of the paper are prohibited.