A Morphological Hessian Based Approach for Retinal Blood Vessels Segmentation and Denoising Using Region-Based Otsu Thresholding

TEAM MEMBERS:

• Aravind Narayanan - 2019102014

• Abhayram A Nair - 2019102017

• Hemant Suresh - 2019102017

• Prayushi Mathur - 2021701034

Goal of the project:

The goal of this project is detection of retinal vasculature by using a morphological hessian-based approach and region-based Otsu thresholding.

Setting up the code:

Install the requirements

pip3 install -r requirements.txt

To view the step-by-step functioning of the segmenation:

View : RetinalSegmentation.ipynb

To open the GUI

python3 script.py

Repository Structure

Folders:

All the data required are present in the repository, however if required they can be downloaded from here as well DATA

1. dataset: 20 input images used for testing the code
2. results: 20 output images of the respeciive inputs
3. testing/labels-ah : 20 ground truth images used for performance analysis
4. resource: contains images used for readme file
5. requirements.txt: Contains list of dependencies(extra dependencies needed to view in jupyter lab are also added)
6. retinalSeg.py: python script version of entire algorithm
7. script.py: GUI implementation
8. RetinalSegmenation.ipynb: Jupyter notebook implementation of algorithm

Link to presentation. Link to presentation pdf

Dataset

The input and ground truth images are obtained from the STARE database of following link: Link The entire unedited dataset obtained from STARE database can be viewed: NOTE: We use the 20 images that are labelled to do the comparison analysis

Sample Inputs:

Rest of the input images used in repo can be found here: Input

Ground truth images: GroundTruth

Sample Output

Rest of the output images can be found here: Results

Algorithm:

View output using graphical User Interface

1. First run the following commmand:

python3 script.py

2. Click on 'browse' button and go into the dataset folder and click 'open'.
3. The list of images would then appear on the left side of the GUI.
4. Then click on any image and wait for a few seconds to see the performance values and the output image.

Demo

View Step by step implementation

The step by step implementation of the code can be viewed on the jupyter notebook named RetinalSegmentation.Ipynb linked here: Link