CS5720 - Week 10
Slide 189 of 200

Face Recognition Systems

Face Recognition Pipeline

Detection
→
Alignment
→
Encoding
→
Matching
Key Insight:
Modern face recognition converts faces into 128-512 dimensional vectors that capture identity while being invariant to pose, lighting, and expression.
Verification vs Identification:
• Verification (1:1): Is this person who they claim to be?
• Identification (1:N): Who is this person from a database?

Key Components

  • 🔍 Face Detection
    MTCNN, RetinaFace, YOLO for finding faces
  • 📐 Face Alignment
    Landmark detection and geometric normalization
  • 🧬 Feature Encoding
    FaceNet, ArcFace, CosFace embeddings
  • ⚖️ Similarity Matching
    Cosine similarity, Euclidean distance, thresholds

Real-World Applications

🔐
Security & Access
Building access, device unlock, border control
📹
Surveillance
Crowd monitoring, person tracking, alert systems
👥
Social Media
Photo tagging, filters, friend suggestions
🛍️
Retail Analytics
Customer tracking, personalized ads, payment
← Previous Next →
Prepared by Dr. Gorkem Kar