High-fidelity dual-image reversible data hiding via prediction-error shift

Abstract Dual-image reversible data hiding (RDH) technology has been an active research area and became an essential part of information security because of its unique advantages in security, embedding capacity and visual quality. For a dual-image RDH strategy, message data is embedded into a cover image to generate two marked images with similar visual quality. In this study, we propose an improved dual-image RDH algorithm based on a prediction-error shift strategy. As a more compact image feature, prediction error is exploited to make better use of the redundancy of image content with the maximum distortion of one. A bidirectional-shift strategy is used to extend the shiftable positions in the central zone of the allowable coordinates. Next, we design a new shiftable-position model for the prediction error corresponding to the peak value in the prediction-error histogram to embed more message bits without causing more distortion. The experimental results demonstrate the efficacy and high-fidelity of the proposed method, especially in relatively low embedding rate circumstances. The average peak signal-to-noise ratio gain of the proposed method on eight standard test images is at least 1.15 dB for the given embedding capacity of 50,000 bits compared with other state-of-the-art dual-image RDH methods.