High-capacity reversible data hiding in encrypted images based on multi-predictions and efficient parametric binary tree labeling

Abstract

Reversible data hiding in encrypted images (RDHEI) enables the embedding of secret data into encrypted images while preserving the ability to fully recover the original images. Existing schemes typically leverage pixel redundancies for data embedding, but they are constrained by the choices of predictors and coding rules, which may result in inefficient bit utilization and increased auxiliary data. This paper presents a novel high-capacity RDHEI method to address these issues. We propose a multi-prediction strategy combining the median edge detector (MED) and the gradient-adjusted predictor (GAP) to improve prediction accuracy. Additionally, we introduce an efficient parametric binary tree labeling approach to categorize image pixels into embeddable, self-recording, and non-embeddable categories, which reduces the generation of auxiliary bits. Experimental results show that our method achieves embedding rates of 3.177, 3.098, 2.722, and 2.6533 bit per pixel (bpp) on the BOSSbase, BOWS-2, UCID, and CT-COVID datasets, respectively, while preserving the security and reversibility of the original image.