Facial Recognition 👨

Authors: Jakub Polczyk, Szymon Leszkiewicz, Kacper Kozaczko

Table of Contents

1. Project description
2. System requirements
3. How to run the app?
4. App Screenshots
5. Experiments results
   1. System's metrices tested on authorized users
   2. System's metrices tested on authorized users and unauthorized users
   3. System's metrices tested on images with Gaussian Noise
   4. System's metrices tested on images wiith luminance noise

Project description

The aim of the project is to use the user authorization system in based on facial biometrics and testing the system's resistance to interference. Project developed in Python, using CelebA dataset and DeepFace model. System should be as secure as possible, therefore it should never accept unauthorized users.

System Requirements

• System design
• Testing the system for user acceptance
• Examine the system for noisy user photos
• Proposing metrics for analysis
• Creation of the User Interface

How to run the app?

streamlit run User_Interface.py

App Screenshots

Welcome Page

Adding user to biometric system

Positive verification

Negative verification

Facial analysis

Verifying multiple users

Experiments results

Experiments performed using

• False Acceptance Rate
• False Rejection Rate
• Accuracy

System's metrices tested on authorized users

FRR	FAR	FAR
0.095	---	0.905

System's metrices tested on authorized and unauthorized users

FRR	FAR	ACC
0.095	0.000	0.953

System's metrices tested on images with Gaussian Noise

Poziom PSNR	FRR	FAR	ACC
70 dB	0.151	0.0	0.927
50 dB	0.131	0.0	0.937
30 dB	0.161	0.0	0.923
20 dB	0.367	0.0	0.823
10 dB	0.889	0.0	0.571

System's metrices tested on images with luminance noise

Poziom zniekształcenia	FRR	FAR	ACC
kwadratowy	0.156	0.000	0.925
liniowy 0.50	0.136	0.000	0.935
liniowy 0.60	0.131	0.000	0.937
liniowy 0.75	0.126	0.000	0.939
liniowy 1.25	0.141	0.000	0.932
liniowy 1.50	0.171	0.000	0.918
o stałą -100	0.312	0.000	0.850
o stałą -20	0.126	0.000	0.939
o stałą -10	0.121	0.000	0.942
o stałą 30	0.131	0.000	0.937