Secure reversible data hiding system in encrypted images for improving payload using efficient parametric binary tree labeling

Recently, an encoding technique called Parametric Binary Tree Labeling (PBTL) has attracted attention in the field of Reversible Data Hiding in Encrypted Images (RDHEI). We observe that the classification strategies for non-embeddable pixels within existing PBTL based methods remain sub optimal, leaving room for further improvement. In order to mitigate these constraints, we develop a novel approach: Improved Efficient Parametric Binary Tree Labeling (IEPBTL). The proposed IEPBTL scheme distinguishes pixels into four independent classes: embeddable pixels, self-recording pixels 1 and self-recording pixels 2, and non-embeddable pixels. Compared with existing methods, IEPBTL significantly reduces the proportion of non-embeddable pixels, thereby enhancing the overall embedding capacity. The findings demonstrate that this refinement effectively curtails the volume of auxiliary data across most benchmarks, which directly contributes to a significant boost in the net payload rate. Furthermore, most existing PBTL based RDHEI schemes adopt a global embedding strategy. Such approaches often overlook the non-uniform distribution of prediction errors in local image regions, thereby failing to fully exploit the available embedding capacity. We overcome the drawbacks of global processing by introducing a preprocessing method that utilizes independent block-based partitioning. For each block, we compute the optimal parameters prior to the data embedding phase. Extensive experimental evaluations confirm that the proposed approach effectively maximizes the embedding capacity while maintaining superior visual quality and reversibility. An enhanced maximum embedding capacity is achieved by the introduced method, providing a significant gain over existing schemes when evaluated against standard test images.