ITR: Theory and Design of Watermarking Codes

University Of Illinois At Urbana-Champaign, Urbana IL

Investigators

Abstract

With the widespread dissemination of digital video, images, audio and other data on the Internet and other media, information protection and copyright protection have become areas of vital importance. These applications have generated an extraordinary level of interest in digital watermarking techniques in recent years. Whereas novel watermarking algorithms and novel ways to defeat them have been developed, fundamental principles of information theory have barely skimmed the surface of this field, and recent results by the PI have shown that existing schemes operate far below the ultimate achievable limits. We intend to develop our ideas for a theory of watermarking codes and to construct codes that approach capacity. We also plan to develop a closely related thrust of research on robust watermarking and authentication techniques for images and video. Our research will be guided by extensive body of knowledge (much of it developed in the last decade) developed in the context of modern communication systems on one hand, and image analysis on the other hand. Our plan is to explore the following topics: Codes for Gaussian channels. We have recently derived closed-form solutions for the hiding capacity of channels involving Gaussian sources and squared error distortion metrics. Our first goal in this research is to develop codes whose performance approaches capacity for such channels. This model, in addition to being useful in the context of practical watermarking applications, will shed light on the fundamental issues involved in constructing watermarking codes. Hence its implications go beyond the particular model studied. Estimation of Attack Channel Parameters. In blind watermarking applications, the decoder does not have access to the original data, and does not know the particular attack that may have been used to corrupt the data. Desynchronization attacks such as scaling, shifting, rotating or warping of image data can then be deadly. We will explore fundamental mechanisms for the decoder to estimate the parameters of such attacks. Codes for arbitrary channels. While Gaussian channels are the worst channels under certain conditions, watermarking codes need to be robust against a variety of attacks. One of our goals is to develop codes and decoding techniques that perform well not only against Gaussian noise attacks, but also against desynchronization attacks, erasures, and other attacks. Application to Image and Video Watermarking. While information theory and coding theory provide fundamental guiding principles to the design of watermarking systems, application of these principles presents unique challenges in specific situations such as image and video watermarking. These two applications will be investigated in detail. Fingerprinting and Authentication Codes. There are several information-hiding problems closely related to watermarking that we intend to explore. Many graduate students are interested in moving into such an attractive research area, which combines breadth and strong emphasis on fundamentals with practical relevance. We intend to train these students for leadership roles in information technology. We also plan to involve undergraduate students, as they find this subject to be a truly enjoyable learning experience.