High-fidelity reversible data hiding using novel comprehensive rhombus predictor

The rhombus mean predictor has been a popular and highly precise predictor commonly deployed for data hiding purposes. However, the rhombus predictor does not always produce the best prediction, for example, when any surrounding pixel is an outlier, because the predictor only calculates the mean of the surrounding pixels without considering their correlation. Therefore, this paper puts forward a comprehensive rhombus predictor (CRP) to take the correlation of the surrounding pixels into account when predicting the centre pixel. CRP adaptively selects the pixels based on their correlation and the characteristics of human visual system for a more precise prediction of the centre pixel. In addition, a highly efficient reversible data hiding (RDH) scheme is proposed using the CRP. The proposed RDH scheme first arranges the pixels in a sequence according to their predicted value by excluding high-complexity pixels. Subsequently, it partitions the sequence into multiple blocks so that the payload can be embedded according to their characteristics by adaptively selecting an embedding strategy. Experiment results demonstrate that the CRP provides higher performance than the existing non-causal related predictors in terms of prediction accuracy. In addition, our RDH based on CRP also outperforms the RDH methods built-upon non-causal related predictors in terms of embedding performance.