Iris image processing to apply Machine Learning algorithms

The purpose of this project is to provide an alternative method to validate the theory of iridology, an alternative medicine techinque that claims to determine information about the health of a person just by analizing the iris. The proposed method consists in extracting segments of iris from eye images, making predictions using a CNN model and then compare the results against the iridology's one.

The CNN model is trained on segments of iris divided in two categories: segments with known problems related to that particular region of the iris (which is related to a particular region of the body according to the iridology's chart) and segments without problems.

If the model is accurate enough we can compare the predictions with the results of an iridology analisys on the same segments. If most of the results differ then we can say that the iridology (for this particular region of the eye) is a false theory. If the entire process is repeated for different parts of the iris than we can say that the iridology is a fake theory.

The project provides a set of scripts to extract the iris segments from the eye images, train the machine learning model and make predictions. The databased used for the tests is the CASIA db and the file config.ini is used to store all the important parameters.

Installation

Clone the repository.

git clone https://github.com/nicoloalbergoni/iridology-CNN

Inside the project directory create a Virtual Environment.

cd iridology-CNN\
virtualenv venv

Then you just need to activate the environment and install all the required packages listed in requirements.txt.

Linux

source /venv/bin/activate
pip install -r requirements.txt

Windows

cd /venv/Scripts/
activate
pip install -r requirements.txt

Usage

Place the eye images related to problems inside the DB_PROBS subfolder contained in the DATA_IMAGE folder wheres the normal labeled images in the DB_NORMAL folder.

Then to start the preprocess task simply run the preprocess.py script.

python preprocess.py

This script performs several operations on the images like: filtering, iris recognition algorithms, segmentation, cropping and resizing, then it saves the extracted segments in a folder called TEMP_SEG. Those segments are rady to be used for the next phase which is training.

To train a model execute the train.py script.

python train.py

The scripts outputs the related .model file and saves it the MODELS folder.

To make predictions copy the desired .model file into the PREDICTIONS directory and add the images on which you want to make predictions to the DATA_TO_PREDICT subfolder. Then run the predict.py script.

python predict.py

The script outputs to the console the predictions results.

Note about the configuration file

The config.ini file contains all the required parameters for the project to work; however the parameters are set to be working right with the selected set of images (CASIA DB).

If you wish you can change the images to work with but you will probably have to also change the value of the parameters in the configuration file to obtain good results. If you don't find the right set of parameters for your images than some images may be discarder due to bad recognition of the iris.