基于支持向量机的眼底图像视盘定位算法  

An Algorithm of Optic Disc Localization in Fundus Image Based on Support Vector Machine

作　　者：吴晔 李成辉 姚骏[2] 

机构地区：[1]中国电子系统工程第四建设有限公司,上海200051 [2]上海大学机电工程与自动化学院,上海200444

出　　处：《工业控制计算机》2023年第2期98-99,101,共3页Industrial Control Computer

摘　　要：视盘定位是计算机辅助诊断中处理眼底图像的重要步骤。为了准确、稳健地确定视盘的位置,提出了一种基于支持向量机(SVM)的视盘定位算法。根据眼底图像中亮区域的结构和强度特征,训练支持向量机分类器识别亮区域。在这些候选区域的基础上创建一个凸包来定位视盘的中心。与文献中的视盘定位方法相比,由于机器学习算法的应用提高了亮区域的分类精度,该方法能够以更高的精度定位视盘中心。测试了三个共259张图像的公共数据集,以评估性能。该方法对DRIVE数据集、DIARETDB0数据集和DIARETDB1数据集的准确率分别为100%、96.9%、97.8%。Optic disk(OD)localization is a significant step when processing the retinal images in computer-aided diagnosis.In order to determine the location of OD precisely and robustly,an OD localization algorithm based on support vector machine(SVM)is proposed in this paper.According to some structural and intensity features of the bright regions in the retinal images,the SVM classifier is trained to recognize bright OD candidate regions.A convex hull is created on the basis of these candidate regions to locate the center of OD.Compared with OD localization methods in literatures,this proposed approach can locate the center of OD with higher accuracy because the application of machine learning algorithm improves the classification accuracy of bright regions.Three public databases with total 259 images were tested to evaluate the performance.The proposed method can achieve an accuracy of 100%,96.9%,97.8%for DRIVE database,DIARETDB0database and DIARETDB1 database respectively.

关 键 词：视盘定位 支持向量机 亮区域 凸包 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程] TP391.41[自动化与计算机技术—控制科学与工程] R77[医药卫生—眼科]