眼底图像中硬性渗出自动检测方法的对比  

Comparative Approaches for Automated Detection of Hard Exudates in Fundus Images

作　　者：高玮玮[1] 沈建新[1] 王玉亮[1] 

机构地区：[1]南京航空航天大学机电学院,南京210016

出　　处：《南京航空航天大学学报》2013年第1期55-61,共7页Journal of Nanjing University of Aeronautics & Astronautics

基　　金：国家高技术研究发展计划("八六三"计划)(2006AA020804)资助项目;中央高校基本科研业务费专项(NJ20120007)资助项目;江苏省科技支撑计划(BE2010652)资助项目;江苏省普通高校研究生科研创新计划(CXLX11-0218)资助项目

摘　　要：为寻求满足临床需求的硬性渗出自动检测方法,从而构建出基于眼底图像的糖尿病视网膜病变自动筛查系统,在利用O tsu阈值分割结合数学形态学快速提取出视盘的基础上,提出了两种硬性渗出自动检测方法(基于数学形态学的硬性渗出自动检测方法和基于RBF神经网络的硬性渗出自动检测方法),在此基础上不仅提出采用后处理以进一步提高检测精度,还就检测结果进行了比较。与其他硬性渗出自动检测方法相比,这两种方法在保证较高检测精度的基础上,效率也较高;在这两种方法之间,基于数学形态学的方法精度更高,基于RBF神经网络的方法效率更高;结合临床对硬性渗出自动检测快速、可靠性的要求,得出基于RBF神经网络的方法作为糖尿病视网膜病变自动筛查系统中的硬性渗出自动检测方法性能更优。In order to establish an automated approach for detecting hard exudates which can meet the clinical requirements, and build the automated diabetic retinopathy screening system, two automatically detecting approaches are proposed, one based on mathematical morphology and the other based on RBF neural network, and they are investigated on the base of segmentation of optic disc with Otsu threshold and mathematical morpholo- gy. Postprocessing is applied to improve the detecting accuracy further. Compared with other approaches in for- mer studies, the two proposed approaches perform well in both accuracy and efficiency of detection. Contrastive analyses between the two proposed approaches show that mathematical morphology-based approach is better in accuracy and RBF neural network-based one is better in efficiency. Considering the request of celerity and dependability in clinic, the approach based on RBF neural network is proposed to be more appropriate for the automated diabetic retinopathy screening system.

关 键 词：眼底图像 硬性渗出 数学形态学 RBF神经网络 自动检测 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]