3.3. Implementation detailsThe same as pairwise PEE [19], the rhombus prediction and the double-layered embedding are adopted. The cover image is divided into “cross” set and “dot” set (see Fig. 8 for illustration), and each set will be embedded with half of the secret message. After rhombus prediction, the pixel-selection (PS) strategy proposed in [19] is employed in our method as well, in which only the prediction-error pairs with complexity less than a preset threshold T will be selected for 2D-PEH generation. Moreover, the treatment of underflow/overflow is also the same as [19], and the details are omitted. For brevity, we only take the first layer to describe the embedding procedure.