Grayscale watermarking resistant to geometric attacks based on lifting wavelet transform and neural network

A blind robust grayscale watermarking scheme that resists geometric attacks is proposed. Firstly, 1-level lifting wavelet transform (LWT) is performed on the cover image and the scrambled grayscale watermarking image, and then the integral wavelet coefficients of the watermarking image are translated into binary bits, which are subsequently embedded into the corresponding frequency domains of the cover image according to perceptual importance. Secondly, to predict the geometric transformation parameters of the attacked image, low-order Tchebichef moments as eigenvectors and an improved back propagation (BP) neural network are utilized to construct the forecasting model, by which the attacked image can be geometrically corrected. Finally the watermarking is extracted from the corrected image. Simulation results show that the proposed scheme is robust to both conventional signal processing and general geometric attacks.