彩色视网膜眼底图像血管自动检测方法

发布时间：2018-03-22 02:05

本文选题：视网膜图像分析　切入点：血管分割　出处：《光学精密工程》2017年05期 　论文类型：期刊论文

【摘要】：为了给视网膜图像配准、光照校正及视网膜内部病理学检测等问题提供有效依据,本文提出一种有效检测及识别彩色视网膜眼底图像血管的全自动方法。针对视网膜可见血管呈长条型管状、局部具有较好直线型结构的形态特点,本文采取适用于条状结构的组合移位滤波响应模型进行特征提取。针对血管和血管末端特征的不同,分别配置对称和非对称的两种滤波模型进行跟踪,利用组合移位滤波模型(对称和非对称)获取到的响应及G通道像素灰度值共同构建特征向量库,采用AdaBoost分类器对各个像素点进行分类判定。基于国际公共数据库DRIVE与STARE的实验结果表明,该方法针对两个标准数据库的分割结果(DRIVE:Accuracy=0.948 9,Sensitivity=0.765 7,Specificity=0.980 9;STARE:Accuracy=0.956 7,Sensitivity=0.771 7,Specificity=0.976 6)均优于已有方法,适用于彩色视网膜眼底图像的计算机辅助定量分析,可作为临床借鉴。
[Abstract]:In order to retinal image registration, illumination correction and retinal pathology detection problem to provide an effective basis in this paper, an effective method of automatic detection and recognition of color retinal fundus images of blood vessels. Blood vessels were visible to retinal long tubular, with local characteristics better linear structure, this paper adopts the combination for strip the structure of shift filter response model for feature extraction. According to the different characteristics of vascular and vascular end, two kinds of filter model are respectively configured for symmetric and asymmetric shift tracking filter model using a combination of (symmetric and asymmetric) construct feature vector library gray scale response and G channel access to the pixel values, using AdaBoost classifier the classification of each pixel is determined. The DRIVE and STARE international public databases based on the experimental results, the method for two standard Quasi database segmentation results (DRIVE:Accuracy=0.948 9, Sensitivity=0.765 7, Specificity=0.980 9; STARE:Accuracy=0.956 7, Sensitivity=0.771 7, Specificity=0.976 6) were better than the existing methods, computer aided quantitative for color fundus image analysis, can be used as a clinical reference.

【作者单位】：吉林大学通信工程学院;长春师范大学计算机科学与技术学院;阿萨巴斯卡大学计算与信息系统学院;
【基金】：国家留学基金委地方合作项目(No.留金法[2013]5045号) 吉林省教育厅“十三五”科学技术研究项目(No.吉教科合字[2016]第001号) 长春师范大学自然科学基金资助项目(No.长师大自科合字[2015]第005号)
【分类号】：R770.4;TP391.41

【相似文献】