Transparent watermarking based on psychovisual properties using neural networks

Abstract

The extreme growth of using digital media has created a need for techniques that can be used to protect the copyrights of digital contents. One approach for copyright protection is to embed an invisible signal, known as a digital watermark, in the image. One of the most important features of an effective watermarking scheme is transparency. A good watermarking method should be invisible such that human eye could not distinguish the dissimilarities between the watermarked image and the original one. On the other hand, a watermarked image should be robust against intentional and unintentional attacks. There is an inherent tradeoff between transparency and robustness. It is desired to keep both properties as high as possible In this paper we propose the use of artificial neural networks (ANN) to predict the most suitable areas of an image for embedding. This ANN is trained based on the human visual system (HVS) model. Only blocks which produce least amount of perceivable changes are selected by this method. This block selection method can aid many of the existing embedding techniques. We have implemented our block selection method in addition to a simple watermarking method. Our results show a noticeable improvement of imperceptibility in our approach compared to other methods.