Iris Recognition

I build an Iris-Recognition system, implemented on both Matlab and Python languages.

Keyword: Iris Recognition, Biometrics, Computer Vision, Image Processing, Daugman

Table of contents

• Iris Recognition
• Table of contents
• I.Introduction
• II.Description
• III.Prerequisites
  • III.1.Matlab
  • III.2.Python
• IV.Folder structure
• V.MATLAB implementation
  • V.1.Enrollment
  • V.2.Verification
  • V.3.Account infomation view
• VI.Python implementation

I.Introduction

• In the last summer 2017, I met my teacher in the course Digital Signal Processing. He recommended me Biometrics topic. Since there, I have started exploring about Biometrics, such as Fingerprint, Iris, and Face.
• I searched on the Internet and found out an open-source Iris Recognition model, which written on Matlab. Thanks to the author of this open-source code, I can build up my own system. Here is information about the author:
  Libor Masek, Peter Kovesi. MATLAB Source Code for a Biometric Identification System Based on Iris Patterns. The School of Computer Science and Software Engineering, The University of Western Australia. 2003.
• Based on this available functions, I have modified, connected, and designed my individual system on Matlab. Subsequently, I have also converted Matlab version into Python one.
• My contribution is creating a GUI so that user can use this as a convenient software. Moreover, in my modified system, I utilized hardware to boost the runtime performance to make is faster, comparing to the original version.
• The testing experiment shows that these two forms had fairly equal accuracy. In addition, because of the C++ platform, Matlab implementation seems faster a little bit than Python.

II.Description

• I notice that this system is not a real-world application. Clearly, a complete system must have a specific camera to capture iris inside eyes. However, these ones are extremely expensive. Therefore, I will use an available image database on the Internet, called CASIA-IrisV1, to replace the costly camera. In this database, there are totally 108 people, each person has 7 eye images. All testing experiments are carried out using images in this database.
• Typically, a recognition system witnesses two operation modes, namely Enrollment and Verification. The former is extracting features from an eye image and save it into a template database. While the latter allows users extract their features and match with existing entities in the template database to identify the origination of the input image, who it belongs to.
• Finally, Matlab version is equiped a familiar GUI for convenient use, whereas, Python version is utilized all CPU cores of hardware for boosting the computation time.

III.Prerequisites

III.1.Matlab

• Because I write the GUI using App designer, and only update versions of Matlab (from R2016a) have this feature. Therefore, to be able to run code, your Matlab version must be R0216a or higher.

III.2.Python

• The OS, which I'm using, is Ubuntu 16.04. In addition, the Python interpreter is Python 3.5.
• First, create a virtual environment, named cv, and install OpenCV library on cv using this link.
• Then, install some python packages inside cv using pip:

workon cv
pip install numpy
pip install scipy
pip install scikit-image

IV.Folder structure

.
+-- CASIA-database/
|   +-- 001_1_1.jpg
|   +-- ...
|   +-- 108_2_4.jpg
|
+-- matlab/
|   +-- fnc/
|       +-- addcircle.m
|       +-- ...
|   +-- template-database/
|       +-- 1.mat
|       +-- ...
|   +-- IrisRecognitionGUI.mlapp
|
+-- python/
|   +-- fnc/
|       +-- boundary.py
|       +-- ...
|   +-- template-database/
|       +-- 1.mat
|       +-- ...
|   +-- path.py
|   +-- enroll-all.py
|   +-- enroll-single.py
|   +-- verify.py

• Folder CASIA-database includes original eye images. My system will use images in this folder as the input.
• Folder matlab is the implementation on Matlab language. In which, folder fnc contains back-end functions for the GUI. Folder template-database stores registered template extracted from eye images. File IrisRecognitionGUI.mlapp is the GUI configuration for my system.
• Folder python is the implemtation on Python language. In which, folder fnc contains back-end functions. Folder template-database stores registered template extracted from eye images. File path.py defines some essential paths. File enroll-all.py is responsible for registering 108 accounts. File enroll-single.py registers for a person, which is indicated by user. File verify.py is used to verify an eye image.

V.MATLAB implementation

• The picture below is the GUI of MATLAB.
• There are 3 modes: Enrollment, Verification, and Account infomation view. Enrollment means register an account template to the database so that you can verify a different eye image belongs to who in the database in the Verification mode. All account templates are anonymous that the name of template file doesn't reveal who owns it. Therefore, Account infomation view can be used to see the information inside a template file.

V.1.Enrollment

• Select button Image select..., a select browser will appear for you to select an eye image for enrollment.

• Then, input the name and basic information that you want to store. Afterward, click button Enroll.

• A message show will notify that the enrollment is successful.

• At this time, I haven't develop a feature to reject accounts that exist in the database. In the future, I will fill into this blank.

V.2.Verification

• To verify an eye image to see who is is, click button Image select... to select an image as instructions in the Enrollment section.
• Click button Verify. Then, a message notify the verification state. In addition, information about the matched account will be shown as the following picture.

V.3.Account infomation view

• Click button View account... to select the template file that you hope to see its content. The result will be same as the picture below.

VI.Python implementation

• First, change the current directory into foler python. Afterward, activate the virtual environment that has been installed OpenCV. Assume that that virtual environment named cv.

cd python/
workon cv

• To register the whole of 108 people in the CASIA database:

python3 enroll-all.py

• To register a specific person:

python3 enroll-single.py 099_1_3.jpg

• To verify a specific person:

python3 verify.py 008_2_2.jpg