Year

2004

Degree Name

Doctor of Philosophy

Department

School of Information Technology & Computer Science - Faculty of Informatics

Abstract

In the era before digital devices, information was in part protected by the cost of duplication. However, as information is more commonly stored to and recovered from digital devices, its duplication becomes trivial. In recent years there has been an increasing amount of research into methods for embedding hidden information into items stored in digital formats. This information can then be used for a range of purposes, including content protection, traitor tracing, authentication and covert communications. In this thesis we exclusively consider the case of information hiding techniques for digital images. We address four problems in this research area: · We consider the question, How much data can be imperceptibly embedded into an image, assuming that the sender and receiver have access to a shared image and that there is no modification of the communication during its transmission? Firstly, the maximum amount of information that can be sent for both monochrome and colour images using this model is derived. We then present coding strategies which nearly achieve the theoretical capacity, operating under the restrictions imposed by image formats, namely that the range of possible pixel values is limited. Results are provided demonstrating that the scheme can be used in practice to achieve the claimed throughput. · We seek discover the set of N codevectors which best represent an n-dimensional, uniformly distributed space. This result could then be considered for the specific case of identifying the general colour palatte in Red/Green/Blue space which is best able to quantize a uniformly distributed colour image. The problem is addressed initially from a theoretical perspective, followed by an empirical analysis. The results are then compared, in terms of distorttion caused by quantization, against lattices, which are known to be an optimal selection of codevectors for large N. · Seppanen, Makela and Kreskinarkaus propose a high capacity steganographic technique to conceal information within a colour image[72], using a shared colour palette. The technique is significant because of the high volume of data that is embedded into pixels but results in a high level of noise and so the quality of the resulting image is not acceptable. We propose: - Technique to improve the tradeoff between capacity and imperceptibility; - New “coding structures” which maintain a high capacity but lower the level of noise; - A strategy for increasing robustness; and - An analysis of the tradeoff by using an alternative algorithm for identifying the colour palette. Results are provided for the evaluation of each proposed modification. The resultant scheme, without robustness, is able to embed an average of over 6 bits per pixel at an imperceptible level of pixel modification. · A robust watermarking technique is proposed which modifies the statistics of extracted segments. By using a region-based approach, resistance is provided to a range of attacks, including rotation, scaling and translation based modifications. The advantage of this algorithm is that no manual pre-processing is required: the detection algorithm is able to confirm the presence of the watermark with only the attacked image. The algorithm is shown to be moderately resistant to a large number of attacks including compression, cropping and rotation

Share

COinS