摘要
为了克服聚类算法对灰度不均匀和有噪声的医学图像分割存在鲁棒性较差等缺点,提出一种基于核密度估计的密度聚类方法分割医学图像。在分析DENCLUE密度聚类算法的思想及爬山策略存在的三个问题的基础上,改进了此密度聚类的爬山策略,并设计了适合于人体组织器官图像分割的DCMIS(Density Clusteringbased Medical Image Segmentation)算法。该算法先用核密度估计数学模型描述医学图像,然后用改进的爬山算法识别聚类,最后根据聚类分割医学图像。该算法有容忍大量噪声数据等特性。实验结果中的欠分割率、过分割率和错误分割率表明DCMIS比DENCLUE和FCM算法有更好的性能和较好的医学图像分割效能。
In order to overcome the bad reliability of clustering algorithm for applications to medi-cal image segmentation led by noise-effect and uneven gray distribution, presents a kernel density estimation based density clustering method to segment medical image. On the basis of analysis of basic ideas of DENCLUE and three main problems of the hill climbing strategy in it, improves the hill climbing strategy for density clustering, and designs DCMIS （ Density Clustering based Medical Image Segmentation） algorithm which is fit for human tissue and organ medial image segmentation. Firstly, DCMIS uses KDE mathematic model to describe medical image data, then takes the improved hill climbing strategy to identify clustering groups, at last segment medical image according to clustering groups. DCMIS has the trait of tolerance for a lot of uniformly distribution noise in data. Experiment results of over-segmentation, under-segmentation and incorrect segmentation rates show that DCMIS has better validity and correctness than DENCLUE and FCM for medical image segmentation.
出处
《计算机应用研究》
CSCD
北大核心
2007年第2期167-169,共3页
Application Research of Computers
基金
国家自然科学基金资助项目(60572112)
镇江市社会发展基金资助项目(SH2003014)
江苏大学第四批学生科研项目医学图像分类系统资助项目
关键词
医学图像分割
核密度估计
密度聚类
爬山算法
Medial Image Segmentation
Kernel Density Estimation
Density Clustering
Hill Climbing Algorithm
