提出了一种基于局部线结构约束的模糊C均值(FCM)聚类眼底视网膜血管分割方法。通过预处理增强血管和背景的对比度信息,采用多尺度匹配滤波器和B-COSFIRE滤波器提取像素特征,然后采用局部线结构约束的FCM聚类算法实现视网膜血管分割,最后通过后处理操作去除孤立的噪声点。在DRIVE数据库的实验结果表明,本文方法的平均准确率为94.21%,平均灵敏度为67.21%,平均特异性为98.2%。与特征空间FCM方法相比,本文方法分割的血管结构的连续性较好,提升了对细小血管检测的灵敏度。

In this study, we propose retinal vessel segmentation based on fuzzy C-means (FCM) clustering in accordance with the local line structural constraints. The pixel features are extracted via multi-scale match filter and B-COSFIRE filter of the pre-processed image, where the contrast between the vessel and the background is enhanced. Thus, retinal vessel segmentation can be realized using the FCM clustering algorithm according to the local line structural constraints. Finally, the isolated noise points are eliminated via the post-processing operation. The experiment is performed using the DRIVE database. The average accuracy, sensitivity, and specificity are 94.21%, 67.21%, and 98.2%, respectively. When compared with the traditional feature-space-based FCM algorithm, the proposed method exhibits better continuity with respect to the segmented retinal vessels and is more sensitive to the small blood vessels.