Authors: Jakub Polczyk, Szymon Leszkiewicz, Kacper Kozaczko
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 design
- Testing the system for user acceptance
- Examine the system for noisy user photos
- Proposing metrics for analysis
- Creation of the User Interface
streamlit run User_Interface.pyWelcome Page
Adding user to biometric system
Positive verification
Negative verification
Facial analysis
Verifying multiple users
Experiments performed using
- False Acceptance Rate
- False Rejection Rate
- Accuracy
| FRR | FAR | FAR |
|---|---|---|
| 0.095 | --- | 0.905 |
| FRR | FAR | ACC |
|---|---|---|
| 0.095 | 0.000 | 0.953 |
| 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 |
| 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 |