Congratulations to Weina Jiang for successfully passing her MPhil to PhD transfer
Wednesday 20 August 2008
Weina successfully transferred on Tuesday 19th August with her work titled " Performance Analysis and Evaluation of Halftone Image Watermarking and Authentication". Weina is supervised by Professor Anthony TS Ho and Dr Helen Treharne and was examined by Dr Chang-Tsun Li (Warwick) and Professor Steve Schneider.
Abstract:
Copyright protection and information security need authentication technologies. Digital watermarking and image hashing are efficient approaches to integrity verification and authentication of images. In the literature, two classes of media verification methods have been developed: watermarking-based authentication and image hashing based signatures. Preserving perceptual quality, robustness and security are the most pronounced aspects for any authentication methods. Our work addresses the performance evaluations of perceptual quality, robustness for watermarking, authentication technologies for halftone images. The unique characteristic of print-scan process causes most watermarking and authentication mechanisms to reduce some of their robustness, particularly in this case, due to the halftoning process and the additional geometric attacks inherent in the scanners. Existing watermarking technologies do not work perfectly for print-scanned image authentication. In this application, a fragile least distortion watermarking approach is proposed and evaluated to resolve the visual quality issue during watermark embedding. Our proposed model can achieve an improvement of between 6.5% to 12% using weighted signal-to-noise ratio (WSNR) and an improvement of between 11% to 23% measured by VIF (visual image fidelity). Moreover, image hashing based on Curvature Scale Space (CSS) descriptor and Analytical Fourier-Mellin Transform (AFMT) are studied for a robust image hashing based authentication of halftone images resilient to geometric attacks. We found the CSS based approach is semi-automatic that requires manual extraction of contour features. It can be used for simple images but not suitable for complex images. However, we found that AFMT based approach is able to differentiate between similar and dissimilar images via the normalised Hamming distance. Initial results indicate that its invariant features are also robust against a number of RST attacks from StirMark and the results are compared.