Signature Recognition: CNN vs. HOG & SIFT Feature Extraction

Overview

This project implements Signature Recognition using Convolutional Neural Networks (CNNs) and manual feature extraction techniques (HOG, SIFT). The goal is to classify signatures based on different individuals and compare CNN-based feature extraction vs. traditional techniques.

---

Key Objectives

• Segment signatures into separate folders per individual
• Perform train-test split for model evaluation
• Train CNN for signature classification
• Compare CNN features with manual feature extraction (HOG and SIFT)
• Evaluate models using Precision, Recall, F1-score, and Accuracy
• Analyze performance through error plots & visualizations

---

Repository Contents

• i201819_B_A1_Q1.ipynb → Jupyter Notebook containing segmentation, feature extraction, and model training
• i201819_ImamaAmjad_Ass1.pdf → Detailed analysis, methodology, and results
• README.md → Project documentation (to be expanded)

For now, please refer to the i201819_ImamaAmjad_Ass1.pdf for dataset details, preprocessing steps, and model evaluation. The README will be expanded soon with additional explanations.

---

Future Enhancements

• Add dataset details & preprocessing steps
• Upload sample outputs & model performance comparisons
• Expand CNN hyperparameter tuning & architecture variations
• Implement additional feature extraction techniques
• Expand the README with dataset details, preprocessing, and architecture explanations
• Add challenges faced and key lessons learned section

---