High-fidelity reversible data hiding technique based on adaptive pairing of prediction errors

Reversible data hiding (RDH) allows the complete recovery of the cover media after data extraction which makes RDH a popular choice among various information sensitive domains. Till date, a plethora of RDH techniques focusing on high-quality marked images with decent embedding capacity have been proposed. Among them pairwise prediction error expansion and pixel value ordering are the noteworthy introductions. However, there is still a requirement of the work which can provide both high-fidelity stego-image along with high-embedding capacity. In this paper, an attempt has been made to achieve better capacity-distortion tradeoff by adaptively choosing prediction error pairs for embedding. An improved version of pairwise PEE has been proposed that hides the secret data into the cover image in two passes by exploiting adaptive pixel pairing. Firstly, the cover image is divided into two independent sets such that alternate pixels are traversed in each set. Then sequential data embedding is performed on the two independent sets where the rhombus predictor is utilized for calculating prediction errors. The pixels in each set are divided into three-pixel blocks such that the pixels in a block are sorted according to their rhombus mean. In the first pass, the first and last pixels are predicted using their corresponding rhombus means, while in the second pass, they’re predicted using the middle pixel. The two prediction errors obtained in each pass are paired together adaptively such that the frequency of embeddable prediction error pairs is increased. It limits the proportion of shifted pixels as well as helps in improving the overall embedding capacity. The experimental results demonstrate that the proposed method achieves reasonable performance improvement when compared with other well-known RDH methods.