A new approach for digital image watermarking to predict optimal blocks using artificial neural networks

Abstract

In this paper, we propose a novel nonblind digital image watermarking based on discrete wavelet transform and singular value decomposition. This robust scheme takes advantage of artificial neural networks for selecting suitable image blocks in which the watermark signal can be embedded. Local characteristics of the blocks such as luminance and texture sensitivity are the main criteria that the selections are based on. Generally, selection is based on a prediction of the results with the objective of transparency and watermark resilience. In other words, before embedding the water mark signal, it is estimated which blocks would be the best for embedding the signals to achieve the desired robustness and quality. Simulation results confirm the superiority of the proposed scheme in terms of the transparency of the images as well as robustness under various kinds of attacks.