一种改进的期望最大图像分割方法

期望最大算法是进行极大似然估计的一种有效方法,它主要用于观测数据不完全或者似然函数不是解析时的参数估计。文中提出了一种期望最大化和贝叶斯信息准则相结合的图像分割方法。首先,运用K均值方法初始化图像分布;然后,运用期望最大算法估计输入图像参数数据,图像中类的数目由贝叶斯消息准则自动确定;最后,运用最大似然标准将像素归类于最相近的类中。实验中将此方法用于对葡萄叶部病害彩色图像的分割,其结果表明此方法有效。

An Improved Expectation-Maximization Image Segmentation Approach

EM algorithm is an efficient tool to deal with the maximum likelihood estimation(MLE).The major applications of EM algorithm is that analysis data is incomplete or the maximum likelihood function is not analytical. The paper proposes an image segmentation approach that is based on Expectation-Maximization and bayesian information Criterion.The Expectation-Maximization theory is used to estimate the data distribution of the input image firstly. The number of class is calculated by Bayesian Information Criterion secondly. The Maximum Likelihood is employed to classify the image pixels into the nearest class finally. This method is used color image segmentation of grape leafage disease. The results show that this method is efficient.