基于超像素仿射传播聚类的视网膜血管分割 
下载： 1123次
Retinal Vessel Segmentation Based on Super-Pixel Affinity Propagation Clustering
华南师范大学物理与电信工程学院, 广东 广州 510006
图 & 表
图 1. ASLICAP方法流程框架图
Fig. 1. Framework diagramfor ASLICAP method
下载图片 查看原文
图 2. 图像预处理。(a)彩色眼底图;(b)绿通道眼底图;(c) CLAHE眼底增强图
Fig. 2. Image pre-processing. (a) Color fundus image; (b) green channel fundus image; (c) CLAHE fundus enhanced image
下载图片 查看原文
图 3. B-COSFIRE 滤波器配置。(a) B-COSFIRE原理图;(b)对称B-COSFIRE;(c)非对称B-COSFIRE
Fig. 3. B-COSFIRE filter configuration. (a) B-COSFIRE schematic; (b) symmetric B-COSFIRE; (c) asymmetric B-COSFIRE
下载图片 查看原文
图 4. 各特征响应图。(a)Hessian最大本征值;(b)Gabor小波变换;(c)B-COSFIRE滤波
Fig. 4. Response mapfor each feature. (a) Hessian maximum eigenvalue; (b) Gabor wavelet transform; (c) B-COSFIRE filter
下载图片 查看原文
图 5. 像素点分类思路图。(a)初始近邻分类;(b) KNN再分类
Fig. 5. Pixel point classification diagram. (a) Initial nearest neighbor classification; (b) KNN reclassification
下载图片 查看原文
图 6. ASLICAP方法在两数据库中的分割图。(a) DRIVE 数据库;(b) STARE数据库
Fig. 6. Segmentation diagrams of ASLICAP method in two databases. (a) DRIVE database; (b) STARE database
下载图片 查看原文
图 7. 同等条件下三种聚类方法分割图。(a)原图;(b)金标准图;(c) ASLICAP; (d) K-means; (e) FCM
Fig. 7. Segmentation diagrams of three clustering methods under the same conditions. (a) Original picture; (b) gold standard; (c) ASLICAP; (d) K-means; (e) FCM
下载图片 查看原文
表 1ASLICAP分割性能指标
Table1. ASLICAP segmentation performance indicators
| Difference | DRIVE database (K=2500) | STARE database (K=4750) |
|---|
| Acc | Se | Sp | | Acc | Se | Sp |
|---|
| Average | 0.9463 | 0.7879 | 0.9725 | 0.9430 | 0.7930 | 0.9581 | | Worst | 0.9358 | 0.7352 | 0.9722 | 0.9203 | 0.6577 | 0.9413 | | Best | 0.9593 | 0.8929 | 0.9711 | 0.9506 | 0.9174 | 0.9551 |
|
查看原文
表 2ALISCAP方法与K-means、FCM算法性能指标对比
Table2. Comparison of performance parameters of ALISCAP method, K-means, and FCM algorithm
| Algorithm | DRIVE database | STARE database |
|---|
| Acc | Se | Sp | | Acc | Se | Sp |
|---|
| K-means | 0.9467 | 0.7515 | 0.9785 | 0.9331 | 0.8184 | 0.9447 | | FCM | 0.9457 | 0.6973 | 0.9852 | 0.9424 | 0.7619 | 0.9609 | | ASLICAP | 0.9463 | 0.7879 | 0.9725 | 0.9430 | 0.7930 | 0.9581 |
|
查看原文
表 3不同算法分割结果比较
Table3. Comparison of segmentation results of different algorithms
| Num | Method | DRIVE database | STARE database |
|---|
| Acc | Se | Sp | | Acc | Se | Sp |
|---|
| 1 | Ref. [7] | 0.938 | 0.781 | 0.966 | 0.887 | 0.767 | 0.939 | | 2 | Ref. [13] | 0.944 | 0.740 | 0.978 | 0.950 | 0.772 | 0.970 | | 3 | Ref.[17] | 0.934 | 0.725 | 0.966 | 0.941 | 0.751 | 0.957 | | 4 | Ref. [18] | 0.937 | 0.703 | 0.971 | 0.932 | 0.758 | 0.950 | | 5 | Ref. [19] | 0.933 | 0.739 | 0.955 | 0.920 | 0.825 | 0.944 | | 6 | Ref. [20] | 0.938 | 0.569 | 0.993 | 0.946 | 0.638 | 0.982 | | 7 | Ref. [21] | 0.947 | 0.780 | 0.972 | 0.945 | 0.769 | 0.938 | | 8 | Ref. [22] | 0.940 | 0.725 | 0.979 | 0.933 | 0.854 | 0.984 | | 9 | Proposed method | 0.946 | 0.788 | 0.973 | 0.943 | 0.793 | 0.958 |
|
查看原文
许言兵, 周阳, 李灿标, 郑楚君, 张润谷, 王文斌. 基于超像素仿射传播聚类的视网膜血管分割[J]. 光学学报, 2020, 40(2): 0210002. Yanbing Xu, Yang Zhou, Canbiao Li, Chujun Zheng, Rungu Zhang, Wenbin Wang. Retinal Vessel Segmentation Based on Super-Pixel Affinity Propagation Clustering[J]. Acta Optica Sinica, 2020, 40(2): 0210002.
PDF全文
