自适应H-minima的改进分水岭堆叠细胞分割方法  

Clustered cells segmentation using modified watershed method based on adaptive H-minima

作　　者：方红萍[1] 方康玲[1] 刘新海[1] 

机构地区：[1]武汉科技大学信息科学与工程学院,武汉430081

出　　处：《计算机应用研究》2016年第5期1587-1590,共4页Application Research of Computers

基　　金：国家自然科学基金资助项目(61105058)

摘　　要：针对传统分水岭方法初始分割结果存在过分割问题,提出自适应H-minima的改进堆叠细胞分割方法。该方法利用不同h值H-minima变换抑制种子噪声,并以对应候选种子为中心,分别采用改进K-均值算法合并初始分割区域,产生候选分割结果;然后,基于形状先验定义圆度指标FuzzyR,并将堆叠细胞平均圆度作为评价函数,自适应提取各堆叠区域最优h值,实现正确分割。实验结果证明,针对于人工合成和真实堆叠细胞图像,算法均能有效抑制过分割、减少欠分割,分割性能显著提高。For the issue of over-segmentation in the watershed initial segmentation,this paper proposed a clustered cells segmentation method based on adaptive H-minima transform in watershed framework. Using corresponding candidate seeds as clustering initial centers,which were reserved seeds after noise seeds suppressed by different h value H-minima transform,this method firstly produced some candidate segmentation results through modified K-means region merging algorithms. Then,it adaptively chose the optimal h value and produced the optional segmentation result for each clustered cells region using maximal average roundness of candidate segmentation results as optimization objective. Experiments on a variety of synthetic and real microscopic cell images show that the proposed method can efficiently restrain over-segmentation and decrease under-segmentation. This method yielded more accurate segmentation rate than the state-of-the-art watershed-based segmentation methods.

关 键 词：分水岭分割 自适应H-minima变换 堆叠细胞分割 K-均值聚类 区域邻接图 类别数目优化 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]