Author Affiliations
School of Physics and Telecommunication Engineering, South China Normal University, Guangzhou, Guangdong 510006, Chinashow less
Fig. 1. Flowchart of the proposed method
Fig. 2. Images of the pre-processing results. (a) Color fundus image; (b) green channel; (c) morphological open operation of Fig. 2(b); (d) image enhancement of Fig. 2(c) by CLAHE
Fig. 3. Multi-scale match filter response images. (a) Response image with σ=1; (b) response image with σ=2; (c) response image with all scales
Fig. 4. Schematic diagram of B-COSFIRE. (a) Principle of B-COSFIRE; (b) symmetrical B-COSFIRE structure; (c) asymmetric B-COSFIRE structure
Fig. 5. Response image of B-COSFIRE filtering. (a) Color fundus image; (b) result of B-COSFIRE filtering
Fig. 6. Schematic diagram of line detector structure. (a) Line detector schematic diagram; (b) schematic diagram of line detector matched with vessel; (c) local neighborhood information
Fig. 7. Segmentation result images of the DRIVE database. (a) The best result of images; (b) the worst result of images; (c) segmentation result of 15th images; (d) segmentation result of 18th images
Fig. 8. Segmentation results of lesion image. (a) Segmentation result of K-means; (b) segmentation result of FCM; (c) segmentation result of proposed method
Fig. 9. Results of proposed method and FCM. (a) Results of FCM; (b) results of the proposed method; (c) segmentation images manually marked by expert
Method | Acc | Sen | Spe |
---|
FCM | 94.04 | 60.77 | 98.98 | FCM_LLC | 94.21 | 67.21 | 98.20 | Difference | 0.17 | 6.44 | -0.78 |
|
Table 1. Segmentation performance comparison of proposed method and FCM%
Method | Average Acc | Average Sen | Average Spe |
---|
The 2nd observer | 94.73 | 77.63 | 97.25 | Chaudhuri et al[7] | 92.84 | 61.68 | 97.41 | Zana and Klein[9] | 93.77 | 69.71 | — | Azzopardi et al[17] | 94.27 | 75.26 | 97.07 | Meng et.al[21] | 93.83 | 58.11 | 93.11 | Kande et al[15] | 89.11 | — | — | Cai et al[22] | 93.00 | 77.00 | 95.00 | Wang et al[23] | 93.82 | 56.86 | 99.26 | Proposed method | 94.21 | 67.21 | 98.20 |
|
Table 2. Performance of different retinal blood vessel segmentation methods%