Improved PPVO-based high-fidelity reversible data hiding.

• An improved PPVO-based reversible data hiding scheme is proposed. • All pixels close to the to-be-predicted pixel are taken as prediction context, contributing to the prediction accuracy enhancement • A new embedding strategy is proposed by conducting multiple histograms generation and modification with multi-sized prediction contexts. • Proposed scheme can be superior to some state-of-the-art RDH methods. Most pixel-value-ordering (PVO) based reversible data hiding (RDH) methods conduct pixel prediction on image blocks, while a recently proposed pixel-based PVO (PPVO) method changes the blockwise prediction in PVO to pixelwise manner and achieves better embedding performance. However, in PPVO, the prediction is not accurate since some pixels close to the to-be-predicted one are not utilized. Moreover, pixels with different local complexity are not fully exploited and the embedding performance of PPVO is not optimized. Thus, to better determine the prediction context as well as full use the image local correlation, an improved PPVO-based RDH method is proposed in this paper. First, to improve the prediction accuracy, a new predictor is proposed in which the prediction context is properly selected. Then, for the optimized performance, a new embedding strategy is proposed based on multiple histograms generation and modification with multi-sized prediction contexts. Experimental results verify that the proposed method is superior to PPVO and some other state-of-the-art RDH methods. [ABSTRACT FROM AUTHOR]