Hierarchical reversible data hiding in encrypted images based on multiple linear regressions and multiple bits prediction

Reversible data hiding in encrypted images (RDHEI) can be used as an effective technique to protect the image content and manage the secret data embedded in the encrypted domain, it is useful for storage and management of the image outsourced to the cloud. In this paper, a hierarchical reversible data hiding with high payload in encrypted images is proposed. The pixels in original image are divided into the multiple linear regressions (MLR) predictable and the no MLR predictable. The MLR predictable pixels are processed to generate two levels label map of pixels by adaptive adjustment of predicted data and multiple bits prediction (MBP). The no MLR predictable pixels are used to generate label map of pixels (LMP) by using simple one bit prediction technique. The additional data produced by compression of three LMPs are embedded into the encrypted images. Thanks to the high correlation between pixels in the clear domain of image, the proposed scheme obtains a high embedding rate. The obtained average payloads are 2.9885 bpp, 3.759 bpp and 3.883 bpp for UCID, BOWS-2 and BOSS dataset, respectively. Extensive experiments including comparison tests, usability tests and ablation tests on some standard test images and three datasets are conducted to testify the effectiveness and usability of the proposed RDHEI, and the experimental results demonstrate that the proposed scheme outperforms the state-of-the-art RDHEI methods in payload on most images, the comparisons of performance show that great improvements have been made in the proposed method.