摘要
大多数身体疾病都可引起眼底图像中血管形状和结构的改变,对眼底图像中血管的有效分割有助于各种疾病的早期发现、诊断和治疗。利用眼底图像分割过程中远距离像素之间会存在相互作用的特点,提出一种基于全连接条件随机场模型的眼底图像血管分割方法。该方法通过计算图像中与目标像素距离较远的像素之间的相互作用,增强了对细长结构连接性的检测能力。将原本用于大面积分割事物的方法成功运用到眼底图像的具有细长结构的血管分割上。采用不同的支持向量机实现对不同像素进行分类及相关参数的自动调节。使用与现有大多数血管分割方法共用的DRIVE和HRF数据库进行测试,结果表明该分割方法有效解决了细小血管处连续性较差、血管融合、断裂等问题。与现有的一些分割方法相比,此法更适合于分割细长结构,对血管的分割效果更接近于专家的手动分割结果,且在灵敏性、特异性等方面都有较大改善。
Most body diseases can cause changes in the shape and structure of blood vessels infundus images. Effective segmentation of blood vessels in fundus images is helpful for early detection,diagnosis and treatment of various diseases. Taking advantage of the interaction between distant pixels inthe fundus image segmentation process, a method of fundus image blood vessel segmentation based on arandom field model with full connection conditions is proposed. By calculating the interaction betweenpixels in the image that are far away from the object pixel, the method enhances the detection capabilityof the connectivity of the slender structure. The technique originally used for large area segmentation wassuccessfully applied to segmentation of blood vessels with slender structures in fundus images. Differentsupport vector machines are used to classify different pixels and automatically adjust related parameters.Using DRIVE and HRF databases that are common to most of the existing vascular segmentation methods,the test results show that the segmentation method effectively solves the problems of poor continuity,blood vessel fusion, breakage and the like at small blood vessels. Compared with some existingsegmentation methods, this method is more suitable for segmentation of slender structures. thesegmentation effect on blood vessels is closer to the expert＇s manual segmentation results, and has greatlyimproved sensitivity and specificity.
作者
郭莹
杨禹惠
GUO Ying;YANG Yuhui(School of Information Science and Engineering,Shenyang University of Technology,Shenyang 110870,China)
出处
《微处理机》
2018年第4期56-64,共9页
Microprocessors
关键词
血管分割
眼底图像
全连接条件随机场模型
结构化输出支持向量机
Blood vessel segmentation
Fundus images
Fully connected conditional random field model
Structured output support vector machine
