How saiyam26/face-recognition-with-opencv works
Project Type and Purpose
• It’s a Python-based, command-line face recognition *pipeline* using OpenCV’s Haar Cascade for detection and LBPH for recognition.
• Purpose: capture face images, train an LBPH model, and perform real-time recognition.
File Structure
• Top-level scripts:
• TakePhotos.py — image capture & dataset creation
• Train.py — model training (LBPH)
• Recognize.py — model inference/real-time recognition
• Classifiers folder:
• haarface.xml — pre-trained cascade for face detection
• TrainedLBPH.yml (generated) — trained LBPH model
• No web/frontend components—this is a CLI/data-processing pipeline.
Extract Insights from README
• Workflow:
a. Run TakePhotos.py → collect ~25 face images per user
b. Run Train.py → produces TrainedLBPH.yml in Classifiers/
c. Run Recognize.py → live camera feed, detect & label faces
• Dependencies: Python 3.6+, OpenCV, NumPy, Pandas.
Main Components & Their Interactions
A. Data Capture Module
• Script: TakePhotos.py
• Uses webcam → detects faces via Haar Cascade (haarface.xml) → stores cropped face images in a local “datasets” folder (implied).
B. Dataset Storage
• Directory for raw images, typically named by user ID/name.
C. Training Module
• Script: Train.py
• Inputs: face images + Haar Cascade
• Process: extracts LBPH features, trains model → outputs TrainedLBPH.yml
D. Model Storage
• Classifiers/TrainedLBPH.yml
E. Recognition Module
• Script: Recognize.py
F. External Libraries
• OpenCV (cv2) – face detection, image capture, display
• NumPy – numerical operations
• Pandas – (optional, for data handling)
Architectural Patterns & Design Principles
• Pipeline pattern: clear separation of capture → train → infer stages.
• Modularity: each Python script is a distinct component.
• Single responsibility: one script per major task.