结合分水岭和互信息的医学图像自适应分割方法  

ADAPTIVE SEGMENTATION APPROACH FOR MEDICAL IMAGES COMBINING WATERSHED AND MUTUAL INFORMATION

作　　者：夏倩倩[1] 宋余庆[1] 刘哲[1] 刘雅婧[1] 

机构地区：[1]江苏大学计算机科学与通信工程学院,江苏镇江212013

出　　处：《计算机应用与软件》2015年第1期221-224,274,共5页Computer Applications and Software

基　　金：教育部博士点基金项目(20113227110010);江苏省普通高校研究生科研创新计划项目(CXZZ11_0575);江苏省博士后科研计划项目(1202037);吉林教育厅"十二五"科学技术研究项目(吉教科合字[2013]第448号);校科研项目(1293000483)

摘　　要：针对现有的改进分水岭算法中对弱边界分割精度不高、计算复杂及梯度谷底阈值不能自适应选取等缺点,提出一个结合互信息的自适应医学图像分割方法。首先通过形态学滤波和高斯平滑滤波来增强图像的边缘区域和抑制图像噪声;然后经过多尺度形态学梯度的谷底填充算法来减少分割区域块数,同时利用基于数学形态学的多元图像边缘检测算法来提取图像的边缘对获得的梯度图像进行修正;通过引入互信息量对填充的阈值进行自适应调整实现控制分割区域的数量,最后实现自动优化分割。实验结果表明,该方法最大程度保留了图像的弱边缘信息,参数选取更加合理,自适应程度提高。We present a mutual information-combined adaptive medical image segmentation method, which aims at following problems of existing modified watershed algorithm： low accuracy in weak boundary segmentation, complicated calculation, and the gradient valley thresholdcaunot be adaptively selected, etc. First, the morphological filter and Gaussian smooth filter are applied for enhancing the edge regionsof the image while suppressing the image noise. Then the valley-filling method based on multi-scale morphological gradient is used to reduce the amount of segmenting regions, meanwhile the multivariate image edge detection algorithm based on mathematical morphology is employedto extract image edges to modify the derived gradient image. Finally by introducing mutual information the adaptive adjustment on the fined threshold is carried out to achieve the control on theamount of segmenting regions, and eventually realises the automatic optimised segmentation. Experimental results indicate that the modified algorithmretains image＇s weak edge information to greatest extent, its parametersselection is more reasonable, and its adaptivity is improved as well.

关 键 词：分水岭算法 形态学滤波 多尺度形态学梯度 多元图像边缘检测 互信息 

分 类 号：TP753[自动化与计算机技术—检测技术与自动化装置]