基于B-COSFIRE和相位一致性图像融合的视网膜血管分割  

Retinalvessel segmentation based on image fusion of the B-COSFIRE and phase congruency

作　　者：郑巧娥 郑楚君[1,2] 黄敏雪 ZHENG Qiao’e;ZHENG Chujun;HUANG Minxue(School of Physics,South China Normal University,Guangzhou 510006,China;School of Electronics and Information Engineering,South China Normal University,Foshan Guangdong 528225,China)

机构地区：[1]华南师范大学物理学院,广州510006 [2]华南师范大学电子与信息工程学院,广东佛山528225

出　　处：《激光杂志》2025年第1期75-83,共9页Laser Journal

基　　金：国家自然科学基金(No.10504008)。

摘　　要：视网膜血管的形态特征在眼底相关疾病的诊断和筛查中有重要的意义,然而,由于视网膜血管尺度形状多样且对比度差异大,细小血管的检测是视网膜血管分割的难点。针对这一问题,提出了一种基于BCOSFIRE(Bar Selective Combination of Shifted Filter Responses)和相位一致性图像融合的视网膜血管分割方法,对预处理后的绿通道图像进行对比度增强,接着提取其B-COSFIRE滤波器响应图和相位一致性响应图,采用小波变换对这两个响应图进行融合得到特征响应图,最后采用特征空间模糊C均值聚类算法实现视网膜血管分割。在DRIVE数据集上进行本方法的视网膜血管分割测试实验,其分割性能中的平均准确率,灵敏度,特异性分别为94.50%,69.63%,98.16%,本方法的分割性能良好,对细小血管具有较高的检测灵敏度,且所分割的血管有较好的完整性。The morphological characteristics of retinal vessels play a significant role in the diagnosis and screening of fundus-related diseases.However,due to the diverse scales and shapes of retinal vessels,as well as large differences in contrast,the detection of fine blood vessels becomes a challenge.To address this problem,retinal vessel segmentation based on image fusion of B-COSFIRE(Bar Selective Combination of Shifted Filter Responses)and phase congruency is proposed.The feature response map was obtained using wavelet transform based image fusion of the BCOSFIRE filtering response and phase congruency of the preprocessed green channel.And then retinal vessel segmentation was implemented by the fuzzy C-means clustering algorithm.The experimental results show that the average accuracy,sensitivity and specificity of the proposed method on DRIVE database is 94.50%,69.63%,98.16%respectively.The proposed method exhibits good segmentation performance,relatively complete segmented vascular network,and improving the sensitivity to detection of small blood vessels.

关 键 词：视网膜血管分割 B-COSFIRE 相位一致性 图像融合 

分 类 号：TN911.73[电子电信—通信与信息系统]