Abstract
In this report, the concept of robust and semi-fragile watermarking is described for copyright protection and authentication of digital images. A number of different transforms and algorithms used for robust and semi-fragile image watermarking are reviewed in detail. Four novel robust and semi-fragile transform based image watermarking related schemes are introduced. These include wavelet-based contourlet transform (WBCT) for both robust and semi-fragile watermarking, slant transform (SLT) for semi-fragile watermarking as well as applying generalised Benford’s Law to estimate the JPEG QF, then adjust the appropriate threshold for improving semi-fragile watermarking technique. The proposed WBCT for robust watermarking are evaluated and compared with two other DWT based algorithms with results achieving high degree of robustness against most non-geometrical and geometrical attacks, while maintaining an excellent perceptual quality. For semi-fragile watermarking, the proposed SLT as a block-based algorithm achieves more accurate for copy and paste attacks with non-malicious manipulations, such as additive Gaussian noise compared with existing DCT-based and PST-based schemes. While for the proposed WBCT method, good performance are achieved in localising the tampered regions, even when the image has been subjected to non-malicious manipulations such as JPEG/JPEG2000 compressions, Gaussian noise, Gaussian filtering, and contrast stretching. The average miss detection rate is found to be approximately 1% while maintaining an average false alarm rate below 6.5%. We also propose the use of generalised Benford’s Law model as an image forensics technique for semi-fragile watermarking. This model can improve the lower tampered detection rate caused by the predetermined threshold in semi-fragile watermarking. The threshold is typically fixed and cannot be easily adapted to different amounts of errors caused by unknown JPEG compression. Our proposed method can adaptively adjust the threshold for images based on the estimated QF by using the generalised Benford’s Law with overall average QF correct detection rate of approximately 99% when 5% of the pixels are subjected to image content tampering, as well as compression using different QFs (ranging from 95 to 65).