Face Recognition System

A complete real-time face detection and recognition system built from scratch with Python, OpenCV and facenet-pytorch. Features:

• Multi-face detection in live video (webcam)
• Automatic face crop capture for training
• Embedding extraction using a pretrained FaceNet (InceptionResnetV1)
• Real-time recognition by comparing embeddings against a template database
• “Known” vs. “Unknown” classification with adjustable threshold

---

📁 Directory Structure

face_recognition_system/
│
├── dataset/                  # Generated at runtime (not committed)
│   └── joao/                 # Training images per identity
│
├── detectors/                # (Optional) Custom detection modules
├── embeddings/               # (Optional) Custom embedding/extraction modules
├── utils/                    # (Optional) Helper functions (logging, I/O, etc.)
│
├── .gitignore                # Ignores venv, dataset, embeddings, caches
├── requirements.txt          # Project dependencies
├── capture_faces.py          # Capture face crops from original frame
├── detect_faces_mtcnn.py     # Real-time detection test with MTCNN
├── extract_embeddings.py     # Extracts and saves embeddings to `embeddings.pt`
├── recognize_faces.py        # Real-time recognition using stored templates
└── README.md                 # This file

---

⚙️ Prerequisites

• Python 3.8+
• Git
• Integrated or USB webcam
• (Optional) CUDA-enabled GPU for faster inference

Dependencies

pip install -r requirements.txt

Your requirements.txt might include:

opencv-python
torch
torchvision
facenet-pytorch
pillow

---

🚀 Installation Guide

1. Clone the repo

   git clone https://github.com/your-username/face_recognition_system.git
   cd face_recognition_system

2. Create and activate a virtual environment

   python -m venv .venv
   source .venv/bin/activate    # macOS/Linux
   .venv\Scripts\activate       # Windows

3. Install dependencies

   pip install -r requirements.txt

4. macOS SSL certificates (if you see SSL errors when downloading weights)

   • Run Install Certificates.command located in /Applications/Python 3.x/.

---

🛠️ Usage

1. Test MTCNN detection

python detect_faces_mtcnn.py

Opens your webcam and draws green boxes around all detected faces.

---

2. Capture training images

python capture_faces.py

• Captures 20 face crops (160×160) from the webcam into dataset/<identity>/.
• Press ESC to stop early.

---

3. Extract embeddings

python extract_embeddings.py

• Loads images from dataset/ and extracts 512-dim embeddings with FaceNet.
• Saves a dictionary of tensors to embeddings.pt.

---

4. Real-time recognition

python recognize_faces.py

• Loads embeddings.pt, computes the average template per identity.
• Detects faces, extracts embeddings and compares to templates.
• Displays identity (green) or “Unknown” (red) with distance score.

Sensitivity threshold
Edit at the top of recognize_faces.py:

threshold = 0.8  # 0.6 = stricter, 1.0 = more lenient

---

🔧 How to Add New Identities

1. Change person_name in capture_faces.py (and create dataset/<name>/).
2. Run the script to capture faces for the new identity.
3. Run extract_embeddings.py again to update embeddings.pt.
4. Now recognize_faces.py will recognize the new name as well.

---

💡 Best Practices & Tips

• Alignment: use facial landmarks (eyes, nose) to align and normalize faces before embedding.
• Cosine similarity: swap Euclidean distance for F.cosine_similarity() for greater robustness.
• Logging: record recognition events to CSV for later analysis.
• GUI: consider Streamlit or Flask for a simple web interface.

---

📄 License

This project is licensed under the MIT License. See LICENSE for details.

---

🤝 Contributing

1. Fork this repository.
2. Create a feature branch: git checkout -b my-feature.
3. Commit your changes: git commit -m "Add my feature".
4. Push to the branch: git push origin my-feature.
5. Open a Pull Request.

---

📫 Contact

João Victor Tavares Esteves
✉️ joaovtesteves2002@gmail.com
🌐 GitHub