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

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.