Abstract: This paper revises the algorithm for 2D histogram-based threshold selection by maximum entropy, and proposes one segmentation schema based on revised maximum entropy. The schema divides 2D histogram into four parts：background, target, noise-interrupted background and target, and maximizes the total entropy of the four parts to determine the optimal threshold for image segmentation. This proposed method has 3 advantages：it contains background and target as possible as it can; it is robust to noise; it doesn’t consider the noise and edge information excessively. Experiments show that the method can retrieve good results.

Key words: 2D histogram, information entropy, noise, image segmentation

摘要： 对基于二维直方图的最大熵选取阈值进行了修正，提出了一种基于修正最大熵的图像分割算法。算法通过将二维直方图分为四部分：背景、目标、受噪声干扰的背景和受噪声干扰的目标，以选取这四部分的信息熵的和最大作为阈值的选取准则。该方法有三个优点：尽可能包括背景部分和目标部分；可以有效地提高对噪声数据的鲁棒性；不过度地引入噪声和边缘信息。实验结果表明，该方法具有较好的图像分割效果。

关键词: 二维直方图, 信息熵, 噪声, 图像分割