Brett T. Hannigan

Adaptive color watermarking

In digital watermarking, a major aim is to insert the maximum possible watermark signal while minimizing visibility. Many watermarking systems embed data in the luminance channel to ensure watermark survival through operations such as grayscale conversion. For these systems, one method of reducing visibility is for the luminance changes due to the watermark signal to be inserted into the colors least visible to the human visual system, while minimizing the changes in the image hue. In this paper, we develop a system that takes advantage of the low sensitivity of the human visual system to high frequency changes along the yellow-blue axis, to place most of the watermark in the yellow component of the image. We also describe how watermark detection can potentially be enhanced, by using a priori knowledge of this embedding system to intelligently examine possible watermarked images.

Using digital watermarks with image signatures to mitigate the threat of the copy attack

In some applications, the utility of an image watermarking system is greatly reduced if an attacker is able to extract a watermark from a marked image and re-embed it into an unmarked image. This threat is known as the copy attack. We develop an image signature scheme to be used with digital watermarks to create an image watermarking system that is more resistant to this attack. We describe the image signature algorithm in detail, and how it may be fused with a digital watermark. We then present preliminary results of our system using an image test set of highly correlated images.

Digital watermarking using improved human visual system model

In digital watermarking, one aim is to insert the maximum possible watermark signal without significantly affecting image quality. Advantage can be taken of the masking effect of the eye to increase the signal strength in busy or high contrast image areas. The application of such a human visual system model to watermarking has been proposed by several authors. However if a simple contrast measurement is used, an objectionable ringing effect may become visible on connected directional edges. In this paper we describe a method which distinguishes between connected directional edges and high frequency textured areas, which have no preferred edge direction. The watermark gain on connected directional edges is suppressed, while the gain in high contrast textures is increased. Overall, such a procedure accommodates a more robust watermark for the same level of visual degradation because the watermark is attenuated where it is truly objectionable, and enhanced where it is not. Furthermore, some authors propose that the magnitude of a signal which can be imperceptibly placed in the presence of a reference signal can be described by a non-linear mapping of magnitude to local contrast. In this paper we derive a mapping function experimentally by determining the point of just noticeable difference between a reference image and a reference image with watermark.

Use of web cameras for watermark detection

Many articles covering novel techniques, theoretical studies, attacks, and analyses have been published recently in the field of digital watermarking. In the interest of expanding commercial markets and applications of watermarking, this paper is part of a series of papers from Digimarc on practical issues associated with commercial watermarking applications. In this paper we address several practical issues associated with the use of web cameras for watermark detection. In addition to the obvious issues of resolution and sensitivity, we explore issues related to the tradeoff between gain and integration time to improve sensitivity, and the effects of fixed pattern noise, time variant noise, and lens and Bayer pattern distortions. Furthermore, the ability to control (or at least determine) camera characteristics including white balance, interpolation, and gain have proven to be critical to successful application of watermark readers based on web cameras. These issues and tradeoffs are examined with respect to typical spatial-domain and transform-domain watermarking approaches.