To better exploit image redundancy, instead of utilizing prediction-errors individually, Ou et al. proposed to consider every two adjacent prediction-errors jointly to generate a 2D-PEH, and then modify the 2D-PEH for data embedding [19]. This method is called pairwise PEE. For its data embedding, the prediction-error sequence (e1 , . . . , eN ) is first transformed into (e1 , . . . , eN/2 ), where ei = (e2i−1 , e2i ) denotes a prediction-error pair. Then, the 2D-PEH is generated by counting the occurrences of prediction-error pairs as