视网膜血管分割算法研究  

Research on Segmentation Method of Retinal Vessels

作　　者：王昕[1] 张歆雅 李颖昉[2] 

机构地区：[1]长春工业大学计算机科学与工程学院,长春130012 [2]长春工程学院计算机基础教学中心,长春130012

出　　处：《吉林大学学报（信息科学版）》2016年第5期587-593,共7页Journal of Jilin University（Information Science Edition）

基　　金：吉林省教育厅"十二五"科学技术研究基金资助项目(吉教科合字[2014]第136号);吉林省科技发展计划基金资助项目(201215127)

摘　　要：针对现有的视网膜血管分割方法对微细血管分割精度低的问题,在多尺度单通道线性追踪(MSLTA:Multi-Scale Single-Channel Linear Track)的图像分割方法基础上,提出了一种新的、有效的视网膜血管分割方法。采用Gabor滤波预处理以增强血管信息,利用MSLTA算法获得最初的血管网络,采用连通域标记的去噪方法,去除图像上的斑点噪声,分割出最终的血管。利用国际上公开的DRIVE(Digital Retinal Images for Vessel Extraction)数据库的视网膜图像进行实验,并与现有的常规算法做了对比。多组实验结果表明,该方法的平均精确度达到95.37%,能很好地保留微细血管。Blood vessel retinal image segmentation has important application value in clinical medical. According to the method of retinal blood vessels segmentation, a new and effective method of retinal vessel segmentation is proposed based on MSLTA （Multi-Scale Single-Channel Linear Track） image segmentation. Pretreatment by Gabor filtering enhances blood vessel information, and obtains initial vascular network algorithm with MSLTA. The method of connected component labeling is used to remove noise in order to remove the speckle noise of the image using the Retinal image of the international public DRIVE （ Digital Retinal Images for Vessel Extraction） database to experiment, and compared with conventional existing algorithms, the experimental results showed that the average precision of this method reaches 95.37% and the micro vessel is also able to be preserved.

关 键 词：视网膜图像 血管分割 多尺度单通道线性追踪 连通域标记 

分 类 号：TN911.73[电子电信—通信与信息系统]