Algorithms for digital watermarking.

摘要

Digital watermarking of multimedia has become an indispensable ingredient in multimedia security. Unlike its ancestors in cryptography, watermarking technology has distinct features that make it an attractive technology for copyright protection and similar applications.; The objective of this thesis is to develop reliable and secure structures for multimedia digital watermarking. Towards this goal, we first introduce a set of robust watermarking algorithms in image and audio based on multiresolution analysis. The image watermarking technique is based on varying the locations of the wavelet extrema to embed data. The audio watermarking algorithms use the time-scale modification for data embedding. All algorithms are robust to common signal processing manipulations.; Next, we focus on reliable structures for watermark decoding. In particular, we study the problem of insertions and deletions that commonly occur in watermark decoding. A set of novel techniques for decoding in the presence of insertions and deletions is proposed. We concentrate on the class of convolutional codes and propose new decoding algorithms in the presence of insertions and deletions derived from the well known Viterbi and sequential decoding algorithms. Moreover, we provide a systematic technique for code construction to maximize the robustness against insertions and deletions.; Finally, we study the security of watermark decoding when the detector is publicly available. We highlight the security gaps of common watermark detectors proposed in the literature. Next, we develop a generalized attack for estimating the detector decision boundary and remove the watermark with minimum distortion. We give different implementations of the generalized attack for the most common detectors. Next, we describe a remedy for this security problem through the use of fractal decision boundaries and we establish the effectiveness of the proposed algorithm in securing watermark public detectors.