Doctor of Philosophy
School of Computer Science and Software Engineering
Piper, Angela, Scalable watermarking for images, Doctor of Philosophy thesis, School of Computer Science and Software Engineering, University of Wollongong, 2010. https://ro.uow.edu.au/theses/3171
Digital watermarking allows the embedding of a signal in multimedia content without affecting its quality or usability. This signal can then be used to identify or confirm the creator or owner, track or prevent unauthorised distribution or verify the integrity of the content, and for a variety of other purposes; making digital watermarking a useful component in the protection of digital multimedia content. With growing demand for both high quality and mobile content, scalable compression has become an increasingly important tool in the distribution of digital content online. It allows content to be scaled to suit a wide range of users and device capabilities in changing network conditions. In particular, resolution scalable compression allows adaptation to different display resolutions, and quality scalable compression caters for different bandwidths. However, for low end devices and low bandwidth connections, the scaling process typically alters the content beyond what a traditional digital watermark is designed to handle. Thus the field of scalable watermarking has emerged to provide digital watermarking algorithms that are suitable for scalably compressed content. In this thesis, resolution and quality scalable watermarking is examined in the context of images, with the aim of developing a watermarking algorithm that is both resolution and quality scalable. Precisely what it meant for a watermarking algorithm to be scalable had not previously been formally defined, and informal descriptions would often focus on a certain desirable property for a scalable watermarking algorithm to the exclusion of some other important property. A definition of a scalable watermarking algorithm is proposed, which considers watermark scalability in terms of two properties that describe the ability of the watermark to both survive high levels of scalable compression and yet still adequately protect all layers of the image. Quantitative measures of these properties are also constructed, to allow the scalability of a watermarking algorithm to be evaluated according to the proposed definition. Although scalable image compression allowed two types of scalability – resolution and quality – prior to the work in this thesis, scalable watermarking algorithms typically provided only one type of scalability or the other (or provided both with the provision that one of the two types be selected at the time of embedding). The problem of creating a single watermarking algorithm that provides both resolution and quality scalability, simultaneously, is considered, using JPEG2000 as a representative scalable compression algorithm. A non-blind spread spectrum based watermarking algorithm is developed, by combining resolution scalable coefficient selection with a human visual system based embedding strength. The resulting watermark provides both detectability and graceful improvement, allowing resolution scaling to 1 256th the original image area, and quality scaling to 1 100th the original file size, exceeding the resolution and quality scalability reported for other scalable image watermarking algorithms. A blind, quantization based algorithm is also developed, that provides detectability and graceful improvement for resolution scaling to 1 1024th the area and quality scaling to 1 100th the size, and additionally maintains an exact match between the candidate and extracted watermarks under scaling. This algorithm is adapted for image authentication, and is sensitive to small changes, including Holliman-Memon and collage attacks, but remains undamaged by JPEG2000 scaling. Previous image authentication watermarks that both tolerate scaling and are secure against Holliman-Memon and collage attacks typically offer only one type of scalability and do not provide graceful improvement.
02Whole.pdf (10580 kB)
Unless otherwise indicated, the views expressed in this thesis are those of the author and do not necessarily represent the views of the University of Wollongong.