Improving visual quality of reversible data hiding by twice sorting

In most literatures on reversible data hiding (RDH), the visual quality of marked images is only assessed by PSNR, and the smoothness-priority-based sorting technique is efficient for improving PSNR. However, modifications in smooth areas are conflict with other criterion of visual quality such as Just Noticeable Difference (JND). To reconcile this contradiction, we propose a twice sorting scheme, in which the pixels are first sorted and divided into several levels with a smoothness criterion, and then sorted twice with JND in each level. According to the sorted order, message bits are embedded into the predicted errors of pixels based on rhombus prediction. Experimental results show that this novel method significantly outperforms previous JND-related RDH schemes on not only PSNR but also SSIM and JND distortion.