基于Student-t分布的混合模型图像分割方法

Mixture Model Image Segmentation Method Based on Student-t Distribution

传统图像分割方法在分割被重尾噪声污染的图像时的分割效果不理想。针对该问题,提出一种基于Student-t分布的图像分割方法。该方法根据像素间的空间关系,计算出其先验概率,使用梯度下降法优化参数,从而最小化误差函数,在参数优化后得到像素点的后验概率值,对像素进行标记以实现图像分割。实验结果表明,在处理被重尾噪声腐蚀的图像时,与传统的K-均值、模糊C-均值等图像分割方法相比,该方法的误分率较低,分割效果较好。

Traditional image segmentation methods can not effectively adapt to the case of image smearing with heavy-tailed noise.And the result of segmentation is not satisfactory in dealing with the image contaminated by heavy-tailed noise.This paper presents an image segmentation method based on Student-t distribution.The method calculates the prior probability according to the spatial relationship between the pixels,uses gradient descent method to optimize parameters so as to minimize the error function.The posterior probability values of pixels are obtained based on the optimal parameters.Image segmentation is realized by marking the pixels.Experimental results show that the misclassification ratio is lower and the performance is better when using the proposed method to deal with the image contaminated by heavy-tailed noise,compared with the traditional K-means and Fuzzy Cmeans（FCM）,etc.