Abstract
A new method for inserting invisible watermark into a digital image is described, which uses two layers of orthogonal transforms including block DCT and two dimensional Fourier transform, pseudorandom data shuffling and a dither modulation technique. Neither the original host image nor the inserted data string is needed in detection and extraction of the watermark. Numerical experiments show that watermark embedded in the proposed way is sufficiently transparent, and robust against various attacks. A desired tradeoff between invisibility, robustness, and information embedding capacity can be achieved by appropriately choosing the system parameters and the candidate coefficients for embedding.
