Reversible watermarking based on position determination and pixel pairs

Reversible watermarking (RW) based on position determination and three-pixel difference is proposed in this paper. The main idea of this paper is to obtain two difference values depending on a pixel pair. To achieve this purpose, for a pixel pair, its one pixel is predicted by the context of this pair to get its predicted value. By this way, we can obtain a three-pixel set containing one pixel pair and one predicted value, and thus obtain two absolute difference values. For a three-pixel set, no modification is allowed to the predicted value. This predicted value along with all the neighbors surrounding one pair constitute a set used for evaluating the intra-pair correlation. The incorporation of the predicted value in this estimation set helps to largely enhance the estimation accuracy. According to the strength of correlation, we determine if this pair is located into a smooth or complex region. When the desired embedding rate is low, we only modify those pairs located in smooth regions while keeping the others unchanged. Therefore, the PSNR (peak signal to noise ratio) value is largely increased. Experimental results also demonstrate that the proposed method is effective.