/CV-Project

Primary LanguagePythonMIT LicenseMIT

Iris Recognition

Description

This system is incomplete for a real-world application. A complete system must have a specific camera to capture the iris of the eye. However, these cameras are very expensive. Therefore available image databases CASIA-IrisV1 and MMU2 are used to replace the costly cameras. All testing and machine learning classification are carried out using images in these databases.

Typically, a recognition system involves two operation modes, namely Enrollment and Verification.

Enrollment is extracting features from an eye image and saving it into a template database.

Verification allows users extract their features and match with existing entities in the template database to identify the origination of the input image.

These system utilizes all CPU cores to boost computation time.

Set Up

These system was deployed and executed on Mac M1 OS and the Python interpreter is Python 3.9.7

First, create a virtual environment

conda create -n iris_vnev python=3.8.4
conda activate iris_vnev

Clone this repository

git clone <repo-name> <dir-name>
cd <dir-name>

Install requirements

cd iris/python
pip install -r requirements.txt

Folder Structure

.
+-- CASIA-database/
|   +-- 001_1_1.jpg
|   +-- ...
|   +-- 108_2_4.jpg
|

+-- MMU2-database/
|   +-- 010101.bmp
|   +-- ...
|   +-- 1000205.bmp

+-- python/
|   +-- fnc/
|       +-- boundary.py
|       +-- ...
|   +-- template-database/  # Created when an image is enrolled
|       +-- 1.mat
|       +-- ...
|   +-- enroll-casia1.py
|   +-- enroll-mmu2.py
|   +-- eval_casia1.py
|   +-- eval_mmu2.py
|   +-- requirements.txt
|   +-- verify.py

  • Folder CASIA-database includes original eye images. These system uses these images as the input.

  • Folder pythoncontains the following folders.

    • Folder fnc, this contains the back-end functions. _

    • Folder template-database stores registered template extracted from eye images, it is created when an image is enrolled.

    • File enroll-casia1.pyenrolls the images in the CASIA1 folder.

    • File enroll-mmu2.py enrolls the images in the MMU2 folder.

    • File eval-casia1.py is used to execute a Machine Learning Classification Algorithm that determines the Maximum fscore, Best eye_threshold and Best threshold for the CASIA1 Database. This process does not have to be repeated but you can re-run to check the following results,

      File eval-mmu2.py is used to execute a Machine Learning Classification Algorithm that determines the Optimal Maximum fscore, Best eye_threshold and Best threshold for the images in the MMU2 Database. This process does not have to be repeated but you can re-run to check the following results.

      File verify.py is used to verify if an image sample can be recognized by the system from the pool of template images. The default threshold used for matching is 0.38.

Usage

To Enroll a set of images

python enroll-casia1.py

In the above example, we are enrolling the CASIA1 Database into the system. This encodes the specially selected images using regex and the outputs are stored in a templates database sub-folder, in the case of this example a CASIA1 sub-folder.

To verify an image

python verify.py --file ../CASIA1/3/003_1_1.jpg --temp_dir ./templates/CASIA1/

In the above example, the file we are trying to verify is an image from the third folder in the CASIA1 Database. We point to the particular file by passing an argument to the --file flag.

Provided all images in the CASIA1 Database have been enrolled to the templates database. We can pass an argument to the --temp_dir flag, pointing to the templates sub-folder where the images were enrolled.

The default threshold used for matching is 0.38.

License

MIT

Alt