retinal-vessel-segmentation
A tool for binarizing retinal blood vessel images.
(Left) Unaltered retinal images; (Center) Binarized results; (Right) Ground truth.
Usage
Retinal images with .ppm extension should be placed in the input folder for processing. After running main.matlab,
the binarized images will be generated inside the output folder.
New file names will generate with the pattern: im0001.ppm -> im0001.output.ppm.
Results
This program was tuned using 20 images from the STARE dataset, a collection of retinal images, to develop and test the process. Using manually determined "ground truth" images created by expert Adam Hoover for the 20 STARE images, the average statistics for the binarized images are as follows:
- True positive rate = 0.89
- True negative rate = 0.67
- Balanced accuracy = 0.78
Created by allytrope and TheFeefster.