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.%In order to provide effective foundation for retinal image registration, illumination adjustment and pathological detection of retina interior and other problems, a fully automatic method of detecting and recognizing blood vessel for color retina fundus images effectively was proposed.Aimed at the state with elongated tubular shape and preferably linear structure in local part of visible blood vessel, combinatorial shifted filter response model that is applicable to strip structure was used for feature extraction.Taking different features of blood vessel and the end of blood vessel into consideration, two types of filtering modes with symmetry and dissymmetry were configured for tracking, feature vector library was established by response acquired from combinatorial shifting filter response model (symmetry and dissymmetry) and G channel pixel value together and each pixel was classified and determined by AdaBoost classifier.The experimental result based on international public database DRIVE and STARE shows that the segmentation result of proposed method on two standard databases (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) is better than existing methods.It is applicable to computer-assisted quantitative analysis of color retina fundus images and can be used as clinical reference.
展开▼
