e.g., the pair A = (3, 2) in the lower triangle region is transformed as B = (0, 2) by (6). As the transformation for other triangular regions are similar to the above example, for brevity, we only describe the embedding details for the lower triangle region of the first quadrant.By this transformation, the mapping direction of rightward remains unchanged, while the mapping direction to the upper-right is converted to upward. Then, the expansion bins for pairwise PEE can be described as a path consisting of pairs (k, 0) of the rectangle region with each k ≥ 0, in which the pairs on the path are expanded to embed data while the pairs upon the path are shifted. More specifically, each pair on the path is expended to right or up to embed 1 bit, and each pair upon the path is shifted to its upper neighbor. Based on this viewpoint, actually, any path in the rectangle region starting at (0,0) can derive a reversible embed- ding scheme, if any pair in the path is the right or upper neighbor of its former in the path. In fact, with such a path, the modification mapping can be defined as follows, any pair (x, y) on the path is expanded to (x + 1, y) and (x, y + 1), any pair (x, y) upon the path is shifted to (x, y + 1), and any pair (x, y) below the path is shifted to (x + 1, y). With this mapping, for any pair in the rectangle region except (0,0), there is only one mapping direction to it (i.e., there is only one pair which is mapped to it), which guarantees the reversibility. For example, Fig. 5 presents one expansion path and its corresponding modification mapping. Thereby, from all these paths in the rectangle region, we can select the optimal one that minimize the embedding distortion.