Face Recognition & Detection System

Python pipeline for face detection and recognition using OpenCV. Haar Cascade for detection, LBPH for recognition.

Quick start: pip install -r requirements.txt → python main.py

---

Two-Stage Pipeline

1. Detection — Haar Cascade locates face regions in each frame using cv2.CascadeClassifier.
2. Recognition — LBPH identifies each detected face using cv2.face.LBPHFaceRecognizer, outputting an ID and confidence value.

Data flow: Camera → Grayscale → Haar Cascade detects faces → LBPH predicts ID per face → Map ID to name via config/users.json → Display with label.

---

Algorithms

Haar Cascade

Pre-trained XML classifier. detectMultiScale slides a window over the image at different scales and applies a cascade of tests to reject non-face regions. Returns bounding boxes (x, y, w, h) for each face.

LBPH (Local Binary Pattern Histogram)

Splits each face into cells, computes Local Binary Pattern per cell, builds histograms, and concatenates into one feature vector per face. At prediction: compares query to stored vectors, returns closest ID and distance. Lower distance = better match.

---

Features

• Enroll faces — Collect training images from camera, train LBPH
• Live face recognition — Real-time identification from webcam
• CLI menu — 3 options: Enroll, Recognize, Exit
• ID→name mapping via config/users.json

Press q to exit camera windows.

---

Project Structure

.
├── main.py              # CLI entry point
├── src/
│   ├── config.py        # Paths, thresholds
│   ├── collect.py       # Camera → face crops → dataset
│   ├── train.py         # Dataset → LBPH model
│   └── recognize.py     # Camera → detection → recognition
├── data/dataset/        # User.{id}.{count}.jpg
├── models/              # trainer.yml (LBPH)
├── config/users.json    # {"id": "name"}
└── assets/              # haarcascade_frontalface_default.xml

---

Run

pip install -r requirements.txt
python main.py

Prerequisites: Python 3.6+, webcam. Grant camera permissions on macOS (System Preferences → Security & Privacy → Camera).

Run modules directly:

python -m src.collect
python -m src.train
python -m src.recognize

---

Dependencies

• opencv-contrib-python — Haar cascade, LBPH, video capture
• numpy — Array operations
• pillow — Image loading in train.py