Table 3. Breakdown of training samples used.
Classifier | Images used | Sub-region used | Training sample type | Testing images for EA |
Blood vessel (G) | 300 | 2789 | Image regions | 1000 |
Microaneurysms (G) | 100 | 2100 | 15×15 sub-images | 1500 |
Haemorrhage (G) | 278 | 1785 | Image regions | 1000 |
Background (L) | 300 | 1750 | 32×32 sub-images | 1000 |
Dark lesion (L) | 278 | 2100 | 32×32 sub-images | 1000 |
Blood vessel (L) | 300 | 4210 | 32×32 sub-images | 1000 |
Bright lesion(L) | 250 | 1889 | 32×32 sub-images | 1000 |
G and L indicate the corresponding classifier is either global or local classifier.