Abstract
With the popularity and affordability of advanced digital image editing software, users can manipulate images relatively easily and professionally. As a result, the task of guaranteeing the authenticity of digital images, particularly as evidence in the court of law, is becoming more important. In the past few years, semi-fragile watermarking has become increasingly important to verify the content of images and to localize the tampered areas, while simultaneously tolerating some non-malicious manipulations. Moreover, some researchers have proposed self-restoration schemes such that the content of tampered areas can be recovered after the authentication process. In this thesis, we introduce three semi-fragile watermarking schemes for image content authentication and self-restoration and one general-purpose digital watermarking benchmarking framework with high reconfigurability. First, we propose a fast watermarking scheme which utilises the relationship between the four 4×4 sub-blocks mean pixel values and the first four DCT coefficients of 8×8 block for image self-restoration. It is evaluated and compared with two other existing watermarking schemes, and achieves an improvement of PSNR approximately 4 dB for the watermarked image and approximately 2 dB for the restoration image. In order to improve the security of the algorithm, we also propose and evaluate an enhanced perceptual image authentication approach by incorporating perceptual hashing into digital watermarking. This approach can provide better robustness to most content-based image processing operations such as JPEG compression and Gaussian noises than most existing semi-fragile watermarking methods. Experimental results achieved also demonstrate a high authentication accuracy rate to non-malicious and malicious image processing operations. This thesis also presents a new image restoration method based on a linear optimization model which restores part of the image from structured side information (SSI). Our experiments show that given complete information of the SSI, the proposed image restoration technique can outperform the state-of-the-art model by approximately 2 dB in terms of average PSNR value and approximately 0.04 in terms of SSIM value. Finally, we present a new OR-Benchmark framework with open interfaces and high reconfigurability that allows the end user to easily extend the functionalities and freely configure the benchmarking process to obtain and analyse the results. This framework is implemented as a MATLAB software package and can be used to benchmark and further improve existing image authentication and self-restoration watermarking algorithms.