Reversible Image Watermarking Using a Diagonal Structure for Prediction

This paper presents a study on the prediction-based reversible data hiding for digital images. By adopting a diagonal structure for prediction, a large number of pixels in the host image can be predicted and further modified for data embedding. Two prediction schemes are derived from the same structure so that the performance with the different numbers of prediction errors can be compared. For either scheme, the optimal prediction mode is found out to minimize the distortion of the host after embedding a certain amount of information. Besides, the host image can be blindly recovered with the overhead information hidden in it. Experiments show that the proposed schemes generally outperform some existing schemes in terms of embedding rate versus distortion.