摘要
眼底视网膜血管图像的纹理与结构信息可作为医学对相关疾病诊断的重要依据。针对视网膜血管存在伪影与尺度结构复杂等难题以及微血管分割较低等问题,提出一种基于多尺度滤波的有监督学习视网膜血管分割算法。采用二维K-L变换综合分析彩色图像三通道的频带信息得到视网膜灰度图像,并利用受限对比度直方图均衡化增强血管与背景的对比度,利用Retinex降低伪影与视盘的干扰;由多尺度高斯匹配滤波、多尺度形态学滤波、Frangi滤波以及2D-Gabor滤波提取相关血管特征,并将提取好的特征集由AdaBoost初步提取血管;利用血管连通域信息去除初分割结果的非血管像素,获得最终的血管图像。该算法在DRIVE与STARE数据集上实验,准确率分别达到96.34%与95.83%。
The texture and structure of the blood vessel image in retinal fundus can be used as important reference for medical diagnosis of related diseases.Aiming at the problems of retinal vessels artifacts,complex scale structure and low microvascular segmentation,we proposed a supervised learning retinal vascular segmentation algorithm based on multi-scale filtering.Using two-dimensional K-L transform,we obtained the gray-scale image of retina by synthetically analyzing the three-channel band information of color image.The contrast limited adaptive histogram equalization was adopted to enhance the contrast between blood vessel and background.We used Retinex to reduce the interference between artifacts and visual discs.Multi-scale Gauss matched filtering,multi-scale morphological filtering,Frangi filtering and 2D-Gabor filtering were used to extract the relevant vascular features.The extracted feature set preliminarily extracted the blood vessel by the AdaBoost algorithm.The non-vessels pixels of the initial segmentation result were removed by using the information of vessels connectivity region,and the final vessels image was obtained.Experiments on DRIVE and STARE datasets show that the accuracy of the algorithm is 96.34% and 95.83% respectively.
作者
梁礼明
盛校棋
蓝智敏
钱艳群
Liang Liming;Sheng Xiaoqi;Lan Zhimin;Qian Yanqun(School of Electrical Engineering and Automation,Jiangxi University of Science and Technology,Ganzhou 341000,Jiangxi,China)
出处
《计算机应用与软件》
北大核心
2019年第10期190-196,204,共8页
Computer Applications and Software
基金
国家自然科学基金项目(51365017,61463018)
江西省自然科学基金面上项目(20192BAB205084)
江西省教育厅科学技术研究重点项目(GJJ170491)
关键词
视网膜血管分割
有监督学习
微血管
多尺度滤波
Retinal vessels segmentation
Supervised learning
Micro-vessel
Multi-scale filtering
