DIGITAL IMAGE WATERMARKING

Abstract

This dissertation presents a blind digital image watermarking technique in discrete wavelet transform domain. The human visual system properties of the host image data is exploited by calculating pixel-wise masking for the wavelet coefficients. Instead of using constant embedding strength for the watermark, the strength is varied adaptively for different wavelet coefficients of the host image data. This is done by the pixel-wise masking of the wavelet coefficients. The multiplicative embedding rule is used at the embedder of the watermarking system, where watermark is adapted to the local image properties. The pixel-wise masking is calculated for all the detail sub-band coefficients in the first level of the wavelet decomposition. At the watermark decoder the two detection techniques are employed. In first detection technique the statistical distribution of the wavelet coefficients is described by the zero-mean generalized Gaussian distribution (GGD). In other detection technique statistical distribution of the wavelet coefficients is described with the Laplacian model. The performance of the watermarking system is evaluated via several simulations. The robustness property of the watermarking system is verified by performing simulations under different image processing operations such as JPEG compression, Gaussian low-pass filtering and Wiener filtering of varying window sizes.