Adaptive multi-predictor based reversible data hiding with superpixel irregular block sorting and optimization

ABSTRACT Reversible data hiding (RDH) is a special class of steganography that is capable of recovering the original cover image upon the extraction of the secret data. The main goal of this paper is to develop different adaptive predictors based on superpixel irregular block sorting. Firstly, a superpixel irregular block and sorting strategy is proposed which is applied to histogram shifting for the first time. Then, a multi-directional edge classification method is proposed, which divides pixels into strong edge pixels, normal edge pixels, and weak edge pixels. Moreover, strong edge pixels and normal edge pixels are further divided into four directions. According to edge classification, the most appropriate adaptive multi-predictor is proposed. Finally, an optimization-based data hiding strategy is proposed. The proposed scheme focuses on constructing a sharp enough histogram. The investigational results demonstrate that the proposed scheme achieves large capacity, high image quality, and low complexity.