A simple four-stages reversible watermarking scheme

This paper proposes a four-stages reversible watermarking scheme based on a new predictor. The eight close neighbors are used to determine two estimated values: one based on the horizontal/vertical neighboring pixels, the other on the diagonal ones. The proposed predictor is computed using an weighted average of the two values. The image is processed in four stages and the weights are automatically determined for each stage. This approach mitigates the negative effect of watermarked pixels on prediction. The proposed scheme outperforms similar watermarking approaches, including the rhombus average and the full context predictor. Experimental results are provided.