Digital audio watermarking

This dissertation describes the design, implementation, and evaluation of an audio watermarking algorithm on the basis of a statistical mathematical model. It shows that even based on contradicting requirements a highly configurable method can be developed suitable for a variety of applications. The design of the basic audio watermarking method, relying on hypothesis testing, is motivated by a detailed discussion of possible application scenarios and corresponding boundary conditions determining algorithmic parameters. Several extensions of the basic algorithm to enable embedding of n bits of information and multiple non-interfering watermarks are presented as necessary for the applicability of the method in certain applications. The performance of the watermarking algorithm judged in terms of quality of the watermarked tracks and the robustness of the embedded watermarks has been improved in several ways: To ensure the quality of watermarked audio signals, the integration of a psychoacoustic model is presented, which can easily be adapted to new developments of more sophisticated models. Improvements on the robustness compared to the basic method are demonstrated both on the watermark encoder and detector. To permit an exhaustive evaluation of the method in terms of quality of the watermarked audio tracks, methods used in the evaluation of high quality lossy compression audio coders have been investigated and adapted to the watermarking problem. Both subjective listening tests and objective measurement standards have been applied to the watermarked tracks in order to verify the obtained quality, which was validated by a comparison to audio tracks compressed via the MPEG 1 Layer III standard. A general approach for a risk analysis of watermarking applications is presented, analyzing possible attacks during the deployment of watermarking methods. This is supported by a categorization of attacks on watermarking systems based on their effect during watermark detection and the assumptions made about attackers. According to this classification, the robustness and the security of the watermarks has been demonstrated by a detailed evaluation of the method with regard to removal, de-synchronization, embedding and false alarms attacks.