Efficient reversible image watermarking by using dynamical prediction-error expansion

Reversible watermarking is a special watermarking technique which allows one to extract both the hidden data and the exact original signal from the watermarked content. In this paper, a recently introduced reversible image watermarking method based on prediction-error expansion is further investigated and improved. Instead of taking the pixels with small prediction-error as embedding pixels (i.e., the pixels that carry watermark bits), we propose to select these pixels in a dynamical way. In fact, we can pre-calculate the embedding distortion for each possible choice of embedding pixels, and determine the one with minimal distortion. We see that, with this choice of embedding pixels, the distortion is reduced comparing with the original method, and thus, the proposed approach has a better performance. In addition, experimental results show that the novel method outperforms some state-of-the-art algorithms.