🧠 Real-Time Face Recognition & Anti-Spoofing System

Version: v1.2.1
FPS: ~12–16 on CPU (buffalo_s)
Accuracy: 95%+ (w/ anti-spoof model)

---

📌 Overview

This is a real-time face recognition system with optional anti-spoofing, built using:

• 🤖 InsightFace for face detection & embedding (SCRFD + ArcFace)
• 🎭 MiniFASNet for spoof detection
• ⚡ Cosine similarity for face matching
• 📦 SQLite for local embedding + match logging
• 🧵 Threaded webcam capture for smooth frame processing
• 🔁 Debouncing to prevent noisy log entries
• 🧑‍💻 Command-line interface for easy control

---

✅ Key Features

• 🧠 Unified face detection + embedding (SCRFD + ArcFace)
• 🎭 Real-time spoof detection via MiniFASNet
• 💾 SQLite logging (name, confidence, timestamp)
• ⏱️ FPS counter, Timer profiler, Logger debouncer
• 🔄 Thread-safe video capture
• 🚀 Train spoof detection model with your own dataset
• 🧩 Modular CLI system (register, capture, test)

---

🔄 Version Timeline

Version	Improvements
v1.0.0	Basic YOLOv8 → ArcFace pipeline + SQLite logging
v1.1.0	Unified InsightFace pipeline (SCRFD + ArcFace)
v1.1.1	Threaded webcam + debouncing + faster DB logging
v1.2.1	Anti-spoofing (MiniFASNet), CLI mode switch, self-training

---

🎮 CLI Usage

python main.py

You'll see an interactive menu:

Mode selection:

1. Face Recognition Only
2. Face Recognition + Anti-Spoofing
3. Exit

Each mode provides options:

🟦 Recognition Mode:

1. Register face
2. Test
3. Go back

🟥 Spoofing Mode:

1. Register face
2. Capture real/fake and train
3. Test
4. Go back

---

🧪 Training Your Own Spoof Detection Model

# In CLI → Spoofing Mode → "Capture real/fake and train"

What it does:

1. Opens webcam
2. You press:
   • r → save real image
   • f → save fake image
3. After capture, it trains MiniFASNet on the dataset

Output: models/anti_spoofing/minifasnet_custom_data.pth

---

🎥 Demo: Real-time Face Recognition with Anti-Spoofing

🗂 Folder Structure

face-recognition/
│
├── config/              # App settings
├── core/                # FPSCounter, Timer, webcam manager
├── db/                  # SQLite, face embedding + log DB
├── modules/             # FaceDetector, Embedder, Matcher, Spoof model
├── trainers/            # Fine-tuning logic (MiniFASNet)
├── utils/               # Drawing, debouncer, etc.
│
├── main.py              # Entry point (CLI menu)
├── run_recognition.py   # Handles recognition mode
├── run_spoofing.py      # Handles spoofing mode
├── face_registration.py # Register face with name
├── collect_dataset.py   # CLI tool to collect real/fake spoof data
├── requirements.txt
├── requirements_local.txt
├── Dockerfile
├── docker-compose.yml
└── README.md

---

📊 Performance Benchmarks

Mode	FPS	Accuracy	Notes
Face Recognition	~12–16	≥ 95%	ArcFace (buffalo_s)
With Spoofing	~6–7	~95% (2-class)	MiniFASNet (fine tuning)

Tested on CPU only (AMD Ryzen AI 7)
~200 real + 200 fake images (per your dataset)

---

📸 Architecture

---

⚙️ Dev Setup

# Install base deps
pip install -r requirements.txt

# Optional: for dev tools
pip install -r requirements_local.txt

---

📝 License

MIT © 2025 Sokritha Yen

---

🙌 Credits

• InsightFace
• MiniFASNet (Silent-Face-Anti-Spoofing)