Multi-dimensional constraints-based PPVO for high fidelity reversible data hiding

Abstract

Reversible data hiding (RDH) based on pixel value ordering (PVO) is an effective information security hiding technique that allows complete recovery of the cover image after data extraction. Current methods mainly adopt PVO-based prediction error histograms to obtain better embedding performance in terms of block prediction or pixel prediction, regardless of the correlation between image textures and different prediction contexts. In this work, we propose a multi-dimensional constraints (MDCs) method to explore prediction mechanism across three dimensions using a pixel-based PVO (PPVO) scheme for high-fidelity RDH. The method involves (a) a local complexity (LC) Constraint that focuses on pixels containing similar local complexity instances with more different quantization intervals, (b) an enclosing location (EL) constraint that discovers the most relevant pixels within the fully enclosed context for prediction error, and (c) an optimal mode (OM) constraint that fuses fully enclosed context adaptively based on optimal search during different subsets of prediction modes. In addition, we introduce the multi-dimensional constraints to determine the prediction context in the PPVO-based RDH method, which helps the prediction mechanism to obtain predictions that are more accurate. The proposed MDCs can be easily applied to existing PPVO-based RDH frameworks. Extensive experiments on test-image data sets show that the proposed method outperforms the current state-of-the-art PPVO approaches.