Facial Recognition System

This repository contains Jupyter notebooks implementing a real-time facial recognition system using Adam Geitgey's face-recognition library along with adversarial poisoning experiments. The project demonstrates how to encode, store, load, and recognize faces, and explores how facial recognition models can be attacked and manipulated.

---

Project Structure

face_recognition

• livefacialrecognition.ipynb Runs real-time facial recognition from a live video stream (e.g., webcam). Detects faces, compares them against known encodings, and displays results.

• load_encodings.ipynb Loads previously stored facial encodings into memory for recognition tasks. Useful for reusing data across multiple sessions.

• save_encodings.ipynb Generates and saves face encodings for known individuals. Encodings are stored in a format suitable for later retrieval.

---

poisoning

• poisoning_attack_static.ipynb Demonstrates how to perturb and poison stored face encodings.

  • Loads existing encodings from the face_recognition folder.
  • Slightly modifies (perturbs) facial encodings to simulate adversarial attacks.
  • Re-saves these poisoned encodings and shows how they can cause misclassification.
  • Provides visualization of how poisoned encodings affect recognition results.

This section highlights the vulnerabilities of facial recognition systems to data poisoning attacks and underscores the need for robust, privacy-preserving methods.

---

Features

• Real-time facial recognition using OpenCV.
• Face encoding and comparison with high accuracy.
• Save and load functionality for encoding persistence.
• Modular notebooks for clear separation between data preparation, real-time recognition, and adversarial experiments.
• Example of a poisoning attack that demonstrates security concerns in biometric systems.

---

Installation & Setup

1. Clone the repository:

   git clone https://github.com/alexstvn/ucd_research_facialrecognition.git

2. Install dependencies (recommended: use a virtual environment):

   pip install -r requirements.txt

3. Run Jupyter Notebook:

   jupyter notebook

---

Usage

1. Use save_encodings.ipynb to generate and store encodings of known individuals.
2. Use load_encodings.ipynb to reload encodings when starting a new session.
3. Run livefacialrecognition.ipynb to test real-time recognition with a live camera.
4. Explore poisoning_attack_static.ipynb to study adversarial modifications and their impact.

---

Research Context

This repository is part of my research experience in privacy, security, adversarial machine learning, and real-time facial recognition as part of my research abroad program at University College Dublin in Ireland. The overall scope of the project was in researching federated learning to mitigate privacy-related attacks in virtual reality systems, particularly in facial recognition. My primary task was to experiement with different facial recognition technologies in a live webcam setting and to demonstrate how poisoning works in that context.

In the attached links, you can find further documentation relating to my research:

• Research Reports: Literature Review and Final Report

• Research Poster Presentation View my Research Poster (PDF)

---

Future Work

• Improve recognition speed and efficiency for large datasets.
• Explore privacy-preserving approaches such as federated learning.
• Investigate adversarial robustness and secure encoding storage.
• Enhance robustness under varying lighting and camera conditions.
• Implement in a virtual-reality system.