Eye-Fundus-Image-Segmentation

Python implementation of bacgkround, fundus, disc and cup segmention on fundus images based on conventional image processing methods.

Approach:

This approach is based on morphological operations. Detailed explanations are made in segmentation_algorithm.pdf. In this projects we had 6 different annotations for each BinRushed fundus images.

The segmentation algorithm can be run via:

python seg_main.py

After images are segmented, original images along with their different segmentations are saved as hdf5 files.

Results (Image from BinRushed database)

Original image:

Image annotated by a clinician:

Result of the segmentation algorithm:

Overlaid image of segmentation and annotation:

Postprocessing:

Segmented images can be further processed (i.e: rescaling, normalizing, etc.) via:

python seg_postprocessing.py

Dataset

Fundus images can be found in Deep Blue Data.