High-fidelity reversible data hiding based on enhanced IPPVO and adaptive 2D histogram modification

Abstract

Reversible data hiding (RDH) based on pixel-based pixel-value-ordering (PPVO) has demonstrated superior embedding performance compared to traditional blockwise PVO methods, owing to its pixel-wise approach. Recently, an enhanced version known as Improved PPVO (IPPVO) has been introduced, incorporating an upgraded predictor and embedding strategy based on multiple histogram generation and modification within multi-sized prediction contexts. Despite these advancements, IPPVO still faces challenges, such as inaccurate due to underutilization of nearby pixels and suboptimal embedding performance resulting from individual modification of prediction errors. To address these limitations and fully exploit local image correlation, this paper introduces an enhanced IPPVO-based RDH method. The proposed method introduces a novel predictor to redefine prediction contexts for improved accuracy. Simultaneously, a multi-layer embedding strategy ensures reversibility and comprehensive pixel utilization. Subsequently, two-dimensional (2D) multi-histograms are generated, accompanied by a novel 2D mapping design method. This method identifies optimal candidates through extensive statistical analysis of actual embedding results. Experimental results demonstrate the superiority of the proposed method over IPPVO and other state-of-the-art PVO-based RDH methods. Notably, the proposed method achieves average PSNR values of 60.67 dB with 10,000 bits and 57.80 dB with 20,000 bits, showing maximum average PSNR enhancements of 1.01 dB and 0.88 dB on USC-SIPI images.