As an extension of [12], pairwise PEE [19] generates a 2D- PEH by jointing adjacent prediction-errors and conduct the data embedding based on 2D-PEH modification with a fixed modification mapping. Its performance is better than the conventional PEE for considering the correlations among prediction-errors. As the results demonstrated in Fig. 10, our method performances better than pairwise PEE, and the improvement is somewhat significant. For example, for the image Pepper, our increase of PSNR is as large as about 2 dB. In all the test cases, only for the smooth image Airplane with very large embedding capacities, our method performs similarly with pairwise PEE. Specifically, according to Tables 1 and 2, for an embedding capacity of 10,000 bits and 20,000 bits, our method outperforms pairwise PEE with an average increase of PSNR by 1.21 dB and 0.95 dB, respectively.