Efficient Pixel Prediction Algorithm for Reversible Data Hiding

Prediction-error expansion (PEE) is an important reversible data hiding technique, which can hide large messages into digital media with little distortion. In this paper, we propose a nearest neighborhood pixel prediction algorithm (NNP) for reversible data hiding algorithms based on Chinese Remainder Theorem (CRT), in which a rhombus prediction is applied in NNP, and prediction errors, the difference between pixels and predictions, are modified to embed data. Further, CRT is utilized to adjust the modification size, thus embedding several bits into one embeddable pixel. Laplacian-like distribution of prediction errors is exploited to achieve a trade-off between embedding capacity and visual quality. Experimental results demonstrate that the NNP achieves better embedding capacity with the same stego image quality than the conventional methods.