🏦 Face Recognition Banking System

A secure, modern banking application with facial recognition authentication using a custom-trained Convolutional Neural Network (CNN).

---

👤 Project Owner

Muhammad Hanzlah
📧 Contact: codewithhanzlah@gmail.com
🎓 Arfa Kareem Technology Incubators - 6 Month Course Final Project

---

🎯 Project Overview

This is a full-stack web application that combines:

• 🧠 Deep Learning (96% accuracy face recognition model)
• 🌐 Web Development (Flask backend + modern frontend)
• 💰 Banking System (complete financial operations)
• 🔐 Biometric Security (passwordless login via face scan)

Key Feature: Face authentication replaces traditional password login, providing a seamless and secure banking experience.

---

✨ Features

🔐 Security Features

• Face Recognition Login: CNN model with 96% accuracy
• Biometric Authentication: No password needed for login
• Session Management: Secure 30-minute sessions
• 3-Attempt Limit: Prevents brute force attacks
• Password Protection: Required for sensitive operations (withdraw, transfer)

💰 Banking Features

• ✅ Check Balance: Real-time balance display
• ✅ Deposit Money: Instant deposits with voice confirmation
• ✅ Withdraw Money: Secure withdrawals (password required)
• ✅ Transfer to Beneficiary: Send money to saved contacts
• ✅ Beneficiary Management: Save and view beneficiaries
• ✅ Transaction History: Complete audit trail with timestamps
• ✅ Voice Announcements: Audio feedback for transactions

🎨 UI/UX Features

• Modern, responsive design (works on mobile + desktop)
• Smooth animations and transitions
• Real-time camera preview
• Confidence score display during authentication
• Voice feedback for key actions
• Professional banking interface

---

🛠️ Technology Stack

Backend

• Framework: Flask 3.0
• Language: Python 3.8+
• Session Management: Flask sessions with secure cookies

Machine Learning

• Framework: PyTorch 2.2
• Model: ResNet18 (Transfer Learning)
• Face Detection: MTCNN (FaceNet PyTorch)
• Accuracy: 96% on validation set
• Training Dataset: 233 authorized + 300+ unauthorized face samples

Frontend

• HTML5: Semantic markup
• CSS3: Modern styling with animations
• JavaScript: Camera access, AJAX, voice synthesis
• WebRTC: Real-time camera streaming

Libraries

• torch & torchvision: Deep learning
• opencv-python: Image processing
• facenet-pytorch: Face detection (MTCNN)
• pillow: Image manipulation
• pyttsx3: Text-to-speech (backend)
• Web Speech API: Text-to-speech (frontend)

---

📁 Project Structure

face_banking_system/
├── app.py                      # Main Flask application
├── config.py                   # Configuration settings
├── requirements.txt            # Python dependencies
├── README.md                   # This file
├── face_recognition_model.pth  # Trained CNN model (96% accuracy)
├── .gitignore                  # Git ignore rules
│
├── utils/
│   ├── __init__.py
│   ├── face_auth.py           # Face authentication (singleton)
│   ├── banking.py             # Banking logic & Account class
│   └── decorators.py          # Route protection decorators
│
├── static/
│   ├── css/
│   │   └── style.css          # Modern banking design
│   ├── js/
│   │   └── script.js          # Frontend logic
│   └── images/
│       └── logo.png           # Bank logo (optional)
│
└── templates/
    ├── face_login.html        # Face scan login page
    ├── welcome.html           # Welcome splash screen
    ├── dashboard.html         # Main banking dashboard
    ├── deposit.html           # Deposit page
    ├── withdraw.html          # Withdraw page
    ├── transfer.html          # Transfer to beneficiary
    ├── save_beneficiary.html  # Save new beneficiary
    ├── beneficiaries.html     # View all beneficiaries
    └── history.html           # Transaction history

---

🚀 Installation & Setup

Prerequisites

• Python 3.8 or higher
• Webcam (for face authentication)
• pip package manager

Step 1: Clone Repository

git clone <your-repo-url>
cd face_banking_system

Step 2: Install Dependencies

pip install -r requirements.txt

Step 3: Add CNN Model

• Place your trained face_recognition_model.pth file in the project root
• This file should be the output from your Google Colab training

Step 4: Run Application

python app.py

Step 5: Access Application

Open browser and navigate to:

http://127.0.0.1:5000

---

📸 How to Use

1. Login with Face

• Open the application
• Position your face in front of the camera
• Click "Scan Face" button
• System detects and authenticates your face
• Shows confidence score (must be ≥85%)

2. Welcome Screen

• Full-screen welcome message
• Voice announcement: "Welcome Muhammad Hanzlah"
• Automatic redirect to dashboard after 3 seconds

3. Banking Dashboard

• View current balance
• Access quick actions (deposit, withdraw, transfer)
• See recent transactions
• Navigate to any banking feature

4. Perform Transactions

• Deposit: No password required
• Withdraw: Password required (enter your configured password)
• Transfer: Select beneficiary, enter amount, password required
• Save Beneficiary: Add new contact with 11-digit account number

5. View History

• Complete transaction log
• Timestamps for all operations
• Filter by transaction type

6. Logout

• Secure logout from any page
• Clears session completely

---

🧠 CNN Model Details

Architecture

• Base Model: ResNet18 (pre-trained on ImageNet)
• Transfer Learning: Froze early layers (conv1-layer3)
• Fine-tuned: Layer4 + custom classifier
• Classes: 2 (Authorized / Unauthorized)

Training Details

• Dataset: 233 authorized faces + 300+ unauthorized faces
• Data Augmentation: Rotation, flip, brightness, zoom
• Epochs: 50 (with early stopping)
• Optimizer: Adam (with filtered parameters)
• Loss: CrossEntropyLoss
• Validation Accuracy: 96%

Face Detection

• Library: MTCNN (Multi-task Cascaded Convolutional Networks)
• Confidence Threshold: 90% for face detection
• Authentication Threshold: 85% for authorization

---

⚙️ Configuration

Edit config.py to customize:

# Authentication
FACE_CONFIDENCE_THRESHOLD = 85.0  # Min confidence for login
MAX_LOGIN_ATTEMPTS = 3

# Account Details
# Sensitive values loaded from environment variables

# Session
PERMANENT_SESSION_LIFETIME = 30  # minutes

---

🔒 Security Features

1. Face Authentication: 96% accurate CNN model
2. Session Management: Expires after 30 minutes
3. Attempt Limiting: 3 failed attempts = lockout
4. Password Protection: Required for withdrawals/transfers
5. HTTPS Ready: Can be deployed with SSL certificate
6. No Credential Storage: Face features extracted on-the-fly

---

📊 Performance

• Model Accuracy: 96%
• False Accept Rate: <3%
• False Reject Rate: <5%
• Face Detection Speed: ~100-200ms
• Authentication Speed: ~300-500ms
• Total Login Time: <1 second

---

🎓 Learning Outcomes

This project demonstrates proficiency in:

• ✅ Deep Learning (CNN, Transfer Learning)
• ✅ Computer Vision (Face Detection, Recognition)
• ✅ Web Development (Flask, HTML/CSS/JS)
• ✅ System Design (MVC pattern, Singleton)
• ✅ Security (Session management, Authentication)
• ✅ User Experience (Modern UI, Voice feedback)
• ✅ Software Engineering (Clean code, Documentation)

---

🐛 Known Limitations

• Requires good lighting for face detection
• Single-user system (designed for one account holder)
• In-memory data storage (resets on restart)
• Requires webcam access

---

🚀 Future Enhancements

• Multi-user support
• Database integration (PostgreSQL/MySQL)
• Mobile app version
• Liveness detection (anti-spoofing)
• Dark mode toggle
• Multi-factor authentication (face + OTP)
• Email notifications for transactions
• API for third-party integrations

---

📜 License

This project is created for educational purposes as part of a 6-month course final project.

---

🙏 Acknowledgments

• Instructor: Sir Saad
• Institute: Arfa Kareem Technology Incubators
• Course: 6-Month Generative AI Course
• PyTorch: For the deep learning framework
• MTCNN: For face detection capabilities
• Flask: For the web framework

---

📧 Contact

Muhammad Hanzlah
📧 codewithhanzlah@gmail.com 🔗 github.com/codewithhanzlah 🔗 linkedin.com/in/codewithhanzlah

---

⭐ If you found this project interesting, please star the repository!