SuperGlue Signature Recognition System

AI-powered signature verification system achieving 98.1% accuracy with state-of-the-art SuperGlue technology.

Confusion Matrix Analysis - Signature Recognition System

Our advanced confusion matrix analysis provides detailed insights into system performance with 98.1% accuracy:

Signature Recognition System confusion matrix with 1.9% error rate

Key Insights from Confusion Matrix:

• True Negatives (450): Rejection of different signatures
• True Positives (115): Acceptance of authentic signatures
• False Positives (2): Minimal false acceptances
• False Negatives (9): Low false rejections

📈 Threshold Optimization Analysis v2.0

Comprehensive threshold analysis revealing 0.30 as optimal threshold for maximum accuracy:

Advanced threshold optimization showing peak performance at 0.30 threshold with 98.1% accuracy

📈 Results Examples

📈 Performance Metrics

Metric	Value
Overall Accuracy	98.1%
Precision	98.3%
Recall	92.7%
F1-Score	95.4%
False Positive Rate
False Negative Rate

🚀 Quick Start

Prerequisites

Python 3.9+
PyTorch
OpenCV
NumPy

Installation

git clone https://github.com/gulcihanglmz/superglue-signature-verification.git
cd superglue-signature-verification
pip install -r requirements.txt

Basic Usage

from match_signatures import verify_signature

# Verify signature pair
result = verify_signature("reference.jpg", "test.jpg")
print(f"Match confidence: {result['confidence']:.3f}")
print(f"Verification: {'VALID' if result['is_match'] else 'INVALID'}")

📁 Project Structure

├── models/                # Neural network models
│   ├── superglue.py       # SuperGlue implementation
│   ├── superpoint.py      # SuperPoint keypoint detector
│   └── weights/           # Pre-trained model weights
├── match_signatures.py    # Main verification logic
├── confusion_matrix_analysis_v2.py  # Performance analysis
├── Report.md            
└── requirements.txt      

📊 Visual Analysis

The system includes comprehensive analysis tools:

• Confusion Matrix: Detailed performance breakdown
• Threshold Optimization: Fine-tuned for best results
• Visual Matching: Keypoint visualization and matching display
• Performance Metrics: Professional reporting and analytics

Detailed Performance Metrics

CLASSIFICATION MATRIX:
                 Predicted
                 Different | Same
Actual Different    450   |   2    (99.6% specificity)
Actual Same           9   |  115   (92.7% sensitivity)

� Sample Verification Results

# Example verification output
{
    "signature_pair": "user_123_sample_01.jpg vs user_123_sample_02.jpg",
    "match_confidence": 0.847,
    "predicted_same": true,
    "ground_truth_same": true,
    "verification_result": "AUTHENTIC",
    "processing_time": "0.68 seconds",
    "keypoints_detected": [187, 203],
    "keypoints_matched": 94,
    "match_ratio": 0.847,
    "security_level": "HIGH CONFIDENCE"
}

---

🤝 Contributing

1. Fork the repository
2. Create your feature branch
3. Commit your changes
4. Push to the branch
5. Open a Pull Request

Star ⭐ this repository if you found it helpful!

References

A Python toolkit for pairwise signature matching using SuperPoint + SuperGlue.
It generates a JSON of match predictions for all signature pairs in your dataset, then visualizes results for inspection.

• SuperGlue Pretrained Network (Matching backbone): https://github.com/magicleap/SuperGluePretrainedNetwork

• SuperPoint & SuperGlue papers for algorithmic details:

  • DeTone, Malisiewicz & Rabinovich, “SuperPoint: Self-Supervised Interest Point Detection and Description”, ECCV 2018.
  • Sarlin et al., “SuperGlue: Learning Feature Matching with Graph Neural Networks”, CVPR 2020.