Multimedia data piracy is a growing problem in view of the ease and simplicity provided by the internet in transmitting and receiving such data. A possible solution to preclude unauthorized duplication or distribution of digital data is watermarking. Watermarking is an identifiable piece of information that provides security against multimedia piracy. This thesis is concerned with the investigation of various image watermarking schemes in the wavelet domain using the statistical properties of the wavelet coefficients. The wavelet subband coefficients of natural images have significantly non-Gaussian and heavy-tailed features that are best described by heavy-tailed distributions. Moreover the wavelet coefficients of images have strong inter-scale and inter-orientation dependencies. In view of this, the vector-based hidden Markov model is found to be best suited to characterize the wavelet coefficients. In this thesis, this model is used to develop new digital image watermarking schemes. Additive and multiplicative watermarking schemes in the wavelet domain are developed in order to provide improved detection and extraction of the watermark. Blind watermark detectors using log-likelihood ratio test, and watermark decoders using the maximum likelihood criterion to blindly extract the embedded watermark bits from the observation data are designed.

Extensive experiments are conducted throughout this thesis using a number of databases selected from a wide variety of natural images. Simulation results are presented to demonstrate the effectiveness of the proposed image watermarking scheme and their superiority over some of the state-of-the-art techniques. It is shown that in view of the use of the hidden Markov model characterize the distributions of the wavelet coefficients of images, the proposed watermarking algorithms result in higher detection and decoding rates both before and after subjecting the watermarked image to various kinds of attacks.