Pixel-based pixel value ordering predictor for high-fidelity reversible data hiding

A pixel-value-ordering (PVO) predictor has been used in reversible data hiding methods to produce high-fidelity marked images when the payload size is moderate. The original PVO generates predictions in a block-by-block manner where only two pixels can be predicted in each block. In this paper, a novel pixel-based PVO (PPVO) is proposed with which each pixel is predicted using its sorted context pixels. The benefit of the PPVO is that the pixels in smooth image regions can be better used to embed data, while most of those pixels are neglected in PVO due to the block constraint. Experimental results demonstrate that the embedding capacity of PPVO is larger than that of PVO, especially for relatively smooth images. The obtained marked image quality of PPVO is also better than that of PVO and other state-of-the-art reversible data hiding methods.