Quantization-Based Watermarking: Methods for Amplitude Scale Estimation, Security, and Linear Filtering Invariance

Watermarking is the process of imperceptibly embedding a message (watermark) into a host signal (audio, video). The resulting signal is called a watermarked signal. The message should introduce only tolerable distortion to the host signal and it should be recoverable by the intended receiver after signal processing operations on the watermarked data. Watermarking schemes based on quantization theory have emerged as a result of information theoretic analysis. In terms of additive noise attacks, these schemes have proven to perform better than traditional spread spectrum watermarking because they can completely cancel the host signal interference, which makes them invariant to the host signal. The existence of good lattices in high dimensions that can be directly and efficiently implemented has made quantization-based schemes of practical interest. Quantization (Lattice)-based schemes are vulnerable to amplitude scale and linear filtering attacks, because these attacks introduce mismatch between the encoder and the decoder lattice volumes. Furthermore, these attacks induce a large amount of distortion with respect to the mean squared error, but do not cause significant perceptual degradations. Such operations on watermarked signals are quite common in many applications. In this thesis we study quantization-based watermarking. We incorporate statistical techniques into quantization-based schemes to build watermarking systems that are robust to amplitude scale and linear filtering attacks. These watermarking systems are applicable in situations where the attacks are unintentional, due to standard signal processing operations. Since traditional quantization-based schemes are not robust to amplitude scale and linear filtering attacks, and due to the frequent presence of these operations in many signal processing applications, we develop amplitude scale estimation procedures for quantization-based watermarking, and construct quantization-based watermarking systems that are robust to linear filtering attacks. The estimation procedures are based on Fourier analysis of the watermarked and attacked signals, and maximum likelihood estimation. The robustness to linear filtering is achieved by applying quantization-based techniques on the amplitudes in the frequency domain. To develop the estimation procedures we first derive probability density function models of the watermarked and attacked data for general host signals. We develop a Fourier-based estimation procedure. It exploits the structure in the probability density function of the watermarked data, due to the encoding process. The approach gives accurate results for high watermark-to-noise ratios, for synthetic as well as real signals. To increase the estimation accuracy for low watermark-to-noise ratios, we develop a maximum likelihood estimation approach. The estimation technique performs well even when there is a mismatch between the probability density function of the host signal and that of the model assumed at the estimator. The estimator also gives accurate results for