In this thesis, new image watermarking techniques are introduced based upon the creation of a proposed frequency ‘deadband’. The ‘deadband’ is a frequency region of zero coefficients, created at the highest frequency band of an image, where its sampling frequency is marginally increased. The creation of the zero region occurs when an image is first up-sampled by a factor L, using image interpolation, filtered (to remove base band copies and any frequency components that would be aliased in the down-sample process) and finally down-sampled by a factor M, with M slightly less than L. It is proposed that this zero region or ‘deadband’ will be used to store watermark information. The reason for this is firstly, that the HVS (Human Visual System) is less sensitive to disturbances at high frequency, so the ‘deadband’ capacity should be higher than an equivalent bandwidth at a lower spatial frequency and secondly, that the original unwatermarked image is easily recoverable from the watermarked image by ignoring the deadband. Restating this idea in a similar way, the original image is untouched by the watermark data and therefore, the process of decoupling the host image and watermark is straightforward. After introducing the basic technique, a standard model of the HVS is applied to the deadband watermarking system and modified to set visual thresholds to the magnitude of each deadband coefficient. Above these coefficients, any watermark data added would yield visually disturbing modifications to the spatial domain. This enables more control over the global class of deadband watermarking methods, since any visual distortion would reduce their usefulness. The question as to how much watermark data can be added to the deadband is investigated. This investigation follows two approaches. The second approach estimates the histogram of the watermark through a novel optimization formulation. In this approach, the statistical properties of the DCT domain signals involved are utilized and not the actual signals themselves. An important feature of watermarks is their robustness. This is their ability to withstand attacks of various forms and still remain in a detectable state afterwards. The attacks could be unintentional but can be expected during the life of the image, such as standard JPEG compression, or intentional and malicious, such as in piracy methods, trying to remove the watermark. The robustness of watermarks in the deadband region is investigated. Whilst watermarking in the frequency domain protects the data from many attacks to which spatial domain watermarks are sensitive, the weaknesses of watermarking in the high frequency region are offset by combining deadband watermarking with other standard low frequency watermarking approaches, to develop hybrid watermarking schemes that try and exploit the higher robustness of the low frequencies and the high data capacity of the high frequency region. Finally, some interesting applications of the deadband region are proposed with suggestions for further research to test the prudence of these suggestions.