Reversible Data Hiding In Images Using Dual Embedding And Pixel Prediction

Abstract

Reversible Data Hiding (RDH) has become an essential research area in digital image security because it allows confidential information to be embedded into digital images while ensuring perfect recovery of both the hidden data and the original cover image. Unlike traditional steganography and watermarking techniques that introduce permanent distortion, RDH guarantees complete reversibility, making it highly suitable for sensitive domains such as medical imaging, military communication, forensic analysis, and digital archiving. Over the years, various RDH techniques such as Difference Expansion, Histogaaram Shifting, and Prediction Error Expansion have been proposed to enhance embedding capacity and minimize distortion [14], [6], [15]. However, achieving an optimal balance between embedding capacity, visual fidelity, and computational simplicity remains a challenge.

This research proposes a dual embedding framework that integrates Prediction Error Expansion (PEE) with Least Significant Bit (LSB) substitution to improve embedding efficiency while preserving high image quality. Pixel prediction using a Median Edge Detector (MED) predictor is employed to reduce prediction errors and minimize embedding distortion. The proposed method is evaluated using Peak Signal-to-Noise Ratio (PSNR), Mean Squared Error (MSE), and Structural Similarity Index Measure (SSIM). Experimental results demonstrate that the proposed technique achieves PSNR values above 57 dB even at higher payload capacities, outperforming traditional histogram shifting and standalone LSB approaches [1], [6], [14]. The findings confirm that the proposed dual embedding approach effectively balances capacity, imperceptibility, and reversibility.