A. LEVEL 1: EMPIRICAL SEARCH OF SIDE INFORMATION Ni et al. [17] employed the highest frequency bin as peak bin regardless of how much payload should be hidden. Latter, considering the Laplace distribution of prediction errors (PEs), Li et al. [31] and Sachnev et al. [19] generally utilized two fixed values at the center region of PE histogram, i.e. 0, 1, as peak bins for data hiding. In 2012, Xuan et al. [27] offered a flexible scheme, which found that combination of those bins with less frequency away from the center region of PE histogram as peak bins rather than those ones located at center region, i.e., 0, 1, might be a better choice in some cases. However, in [27], to control the computation complexity, peak bins are mandatory to be continuously selected, which means the case with continuous values, i.e., 2, 3, 4 rather than the combination of random determined values, i.e., 2, 4 , is allowed to be peak bins, even the latter one might be better. After-wards, Hwang [20] incorporated above-mentioned flexible side information selection method in Xuan et al. [27] and Sachnevet al.’s prediction [19] to offer an improved scheme. Based on the above analysis, it is clear that those empirical methods commonly add some additional constraints to achieve a rapid search, which is hardly to achieve global optimal side information.