一种基于区域的交互式MRF图像分割算法设计与实现

摘要

提出一种基于区域的交互式MRF图像分割算法,有效地解决了传统MRF方法中初始参数不易准确估计的问题,并能自动检测和矫正出现的误分割.算法在meanshift预分割得到的过分割区域上进行人工标记,并将图像转化到L*u*v空间,用较简单的欧式距离表达不同像素点之间的差异.同时,分析了误分割可能出现的情况,分类别对误分割现象进行自动检测,然后运用基于区域势能的区域合并方法对误分割进行自动矫正.通过对医学图像及自然彩色图像进行实验,实验结果表明该算法能准确地完成分割任务,并实现对误分割现象的自动检测和矫正,与传统MRF及GraphCut算法相比,该算法能获得更准确及更平滑的分割结果.%An interactive MRF image segmentation algorithm based on region is proposed,which effectively solved the prob-lem that the initial parameters are difficult to accurately estimate in the traditional MRF method,and can automatically detect and correct the false segmentation. The over-segmentation regions obtained by meanshift pre-segmentation are marked artificially with the algorithm. The image is convert into L*u*v space,and the difference of different pixels is expressed with the simple European distance. The situations possible to occur the false segmentation are analyzed,and the false segmentation phenomenon is detected automatically with classification. And then,the region merging method based on the region potential energy is used to correct the false segmentation automatically. The experiments of medical image and natural color image were performed. The experimental results show that the algorithm can accurately accomplish the segmentation task,detect and correct the false seg-mentation phenomenon automatically. In comparison with the traditional MRF and GraphCut algorithms,this algorithm can ob-tain more accurate and smoother segmentation results.