基于分割图像最大信息量的阈值化算法

Thresholding Image Based on Maximum Segmented Image Information

简述了目前利用熵的原理对图像进行阈值化分割的几种方法，把图像假设为Ｌ 个符号的信息源，图像的分割类比于数据信号通过有扰信道过程，分割后的图像具有 一定信息量，提出使分割后的图像具有最大信息量的阈值化方法。将图像背景与目标 的条件概率假设为正态分布，利用贝叶斯公式估计出其后验概率，搜索使分割后的图 像具有最大信息量的阈值，比较了新算法与其它基于香农熵算法的特点和分割性能。

The authors assume in the paper that an image is an L (gray-level) symbol source, that the image segmentation process can be interpreted as data processing that operates on a gray-scale image and produces a segmented image, and that a segmented image contains a certain amount of information entropy, which can be defined as segmented image information. A new algorithm based on maximum segmented image information is presented for thresholding real images. The conditional probabilities of objects and background are assumed in normal distribution and posterior probabilities are computed by Bayes formula. The new thresholding criterion possesses better properties as compared with the maximum entropy (ME) and minimum cross entropy (MCE) thresholding criteria.