一种基于高斯混合模型与Markov建模的灰度图像分割方法

A gray image segmentation method based on Gaussian mixture model and Markov modeling

当灰度图像中存在区域间灰度变化不明显或者含有噪声时,图像分割效果会受到比较严重的影响,本文针对此类问题,基于高斯混合模型,提出了一种改进的灰度图像分割算法。首先,基于马尔可夫随机场建模,将梯度因素引入邻域约束,推导图像的能量函数。然后,采用改进的期望最大(EM)算法对能量函数进行求解,E步通过图割法求解各像素点的分类,M步通过改进的期望最大(EM)算法求解高斯混合模型中的各参数。实验结果表明,本文的方法相对于直接用图割法能够求得更低的能量值,获得较好的分割结果。

When the gray variation among different regions of a gray image is not distinct or there are some noises,the results of image segmentation will be seriously influenced.In this paper,we present an improved gray image segmentation method based on the Gaussian mixture model to solve this problem.First,we introduce the factor of gradient to the neighborhood constraint based on the Markov random filed（MRF） model,and minimize the energy function of the image.Then,we adopt the improved EM algorithm to solve the energy function： E step is to solve the classified problem of each image pixel through graph cut method,and M step is to solve the parameters of Gaussian mixture model through the improved EM algorithm.As the experiment shows,compared with that directly using graph cut method,our method can get a lower energy value and a better result of segmentation.