As a generalization of one-dimensional minimum-error threshold segmentation method was widely used in image segmentation. In this paper, two-dimensional minimum-error threshold algorithm and its fast algorithm (here referring to the original method) are improved, and threshold acquirement formula of the original minimum-error threshold segmentation method was deformed. By using look-up table method, acquirement of two-dimensional threshold is transformed into that of two one-dimensional threshold, so the calculation is simplified and time complexity is reduced from O(L2) to O(L). Experimental results show this decomposition algorithm outperforms original algorithm. Without losing the robustness to noise and needing more time and space, it can get the same segmentation threshold as the original method.