Optimal PPVO-based reversible data hiding

The optimal PPVO-based reversible data hiding (RDH) including adaptive prediction pattern and optimal bin selection is proposed in this paper. By reasonably designing the prediction pattern, each to-be-embedded pixel can be predicted by the n (n∈{4,…,13}) neighbors surrounding it. The larger n, the more accurate the prediction, and the higher embedding performance is achieved at low embedding capacity (EC) and vice versa. Since the neighbors surrounding each to-be-embedded pixel can be used for measuring the local complexity, the smoothness classification is more accurate, and furthermore, the prediction performance is increased. Two-stage embedding is exploited in this paper so as to ensure the implementation of full-enclosed-based prediction. Instead of treating separately two stages like in existing methods, two stages are treated jointly to optimize the selection of embedding bins at a given EC. In this way, the optimal embedding bins which can achieve as high visual quality as possible is obtained. Extensive experiments verify the proposed method is effective.