Table 2. OD boundary delineation results.
| TPR | FPR | Ac | % Overlap | Dice | |
|---|---|---|---|---|---|
| DRIVE (N = 40) | |||||
| MS | 0.882 | 0.003 | 0.993 | 81.0 | 0.881 |
| G | 0.851 | 0.004 | 0.992 | 78.3 | 0.855 |
| R | 0.526 | 0.014 | 0.974 | 48.3 | 0.525 |
| MESSIDOR (N = 1,200) | |||||
| MS | 0.876 | 0.003 | 0.994 | 80.2 | 0.875 |
| G | 0.770 | 0.001 | 0.993 | 72.5 | 0.820 |
| R | 0.772 | 0.006 | 0.988 | 70.1 | 0.759 |
Notes:
Results are shown for the two public databases, DRIVE and MESSIDOR. The number in brackets, N, is the number of images in the database. The three rows per database, show the results for the MS (multispectral), G (only green) and R (only red) test. The columns show metrics derived from the contingency tables for the classification of pixels as OD versus the expert marked OD. TPR, the true positive rate; FPR, the false positive rate; Ac, the accuracy; % Overlap, the percentage of overlap; Dice, the Dice index. The bold entries indicate the best performance.