Reversible Data Hiding in Encrypted Images Based on Edge-Directed Prediction and Multi-MSB Self-Prediction

Abstract

With the development of cloud computing, reversible data hiding in encrypted images (RDH-EI) technology has gained significant attention in ensuring data security and privacy protection. This paper proposes a high-capacity RDH-EI method that employs a dual prediction strategy, including edge-directed prediction and multiple most significant bits (multi-MSB) self-prediction. First, the image owner utilizes the edge-directed property of least-square optimization to enhance the accuracy of pixel prediction. Subsequently, each pixel undergoes self-prediction starting from the MSB and progressing towards an adaptive high-order bit-plane, plane by plane. A binary location map marks the prediction errors for each bit-plane. After being losslessly compressed using the Joint Bi-level Image Experts Group (JBIG) algorithm, the location maps are embedded into the encrypted image as auxiliary data, thereby creating ample room for embedding additional data. Next, the data hider retrieves the auxiliary data from the encrypted image and embeds the additional data into the reserved embedding room, resulting in a marked encrypted image. Finally, authorized recipients with different keys can either accurately extract data or precisely restore the image from the marked encrypted image without any errors. The experimental results show that the proposed method achieves average pure embedding rates of 4.240 bpp for BOSSBase dataset and 4.088 bpp for BOWS-2 dataset, respectively, outperforming many state-of-the-art RDH-EI methods.