Doctor of Philosophy (PhD)
Computer Science Department - Faculty of Informatics
Kailasanathan, Chandrapal, Securing digital images, PhD thesis, Computer Science Department, University of Wollongong, 2003. http://ro.uow.edu.au/theses/174
With the advent of the Internet and World Wide Web, images are being transferred over insecure channels from different parts of the world. Those who receive images may not exactly know where they have originated from or who their sender was. In those circumstances, determining the authenticity of received images would be extremely difficult and therefore providing authenticity to images has been considered as an essential area of image security research. On the other hand, processing images for storage reduction, noise removal, or quality enhancement, is also essential in many computer applications, because images are inherently larger than text and are subject to noise or transcoding. The authentication of image is achieved either by appending a Message Authentication Code or a Signature to an image or by embedding a watermark into the image which is fragile to any modification. On the other hand, storage reduction for images is mainly achieved by lossy compression techniques such as JPEG or MPEG, and the quality enhancement or noise reduction for images is accomplished by employing filtering techniques such as Median or Gaussain. In this thesis, we propose hash functions which extract essential image features that survive acceptable level of JPEG compression while distinguishing other modifications using statistical measures of central tendency, dispersion, hypothesis of testing, and k-mean clustering, and wavelet/discrete cosine transform. For the proposed transform based hash functions, we optimise performance by fixing some essential parameters such as the size of hash, the value of the threshold and the low-pass filter used. We also propose two oblivious fragile watermarking schemes. The first one embeds a logo on feature points or critical point extracted using wavelet or discrete cosine transform, and the next one embeds a logo on image blocks based on polarity of pixel points. For the proposed schemes, after analysing the security level, the possibilities of embedding multiple logos for secret sharing have also been investigated. The appendix of the thesis addresses some of our early work on combined compression and encryption scheme based on adaptive Huffman coding, compression performance of a proposed JPEG encryption scheme, and cryptanalysis of Huffman coding with four symbol alphabet.