基于多变量类别自适应的图像分割算法  

Image Segmentation Based on Class Adaptive Spatially Variant Mixture Model

作　　者：常兴治[1] 高立群[2] 

机构地区：[1]中国科学院青岛生物能源与过程研究所,青岛266071 [2]东北大学信息科学与工程学院,沈阳110004

出　　处：《计算机辅助设计与图形学学报》2008年第10期1383-1388,共6页Journal of Computer-Aided Design & Computer Graphics

基　　金：公安部重点项目(20029322301);黑龙江省自然科学基金(F0318)

摘　　要：提出一种根据分割要求自动配置区域平滑性的图像分割算法.通过对现有的类别自适应空间变量混合模型进行改进,修改模型算法中Markov随机场的势函数部分,在判断分割区域环节时引入了像素的色彩或灰度信息,使改进后算法的稳定性有明显提升;同时增加了像素强度系数α以及算法的灵活性,提高了其实用价值.最后在MIT及Berkeley的分割测试图片上进行仿真实验,证明了该算法的有效性.This paper proposes an image segmentation method which allocates regional smoothness automatically according to the requirements. The method improves existing class adaptive spatially variant mixture model based on revising the potential function in Markov random field and introducing colorful or grey feature information into the assessment of segmentation regions. As a result, the stability by the revised method is improved significantly. Meanwhile, the added pixel strength coefficients to the method increase its flexibility and practical value greatly. Finally, the algorithm efficiency is approved through experimental simulation on the test images from MIT and Berkeley galley.

关 键 词：MARKOV随机场 图像分割 EM聚类 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]