This project focuses on building an Offline Signature Verification System that utilizes a Siamese Neural Network to authenticate signatures and YOLOv8 to extract signatures from documents. The system automates the verification process, which is often manual, time-consuming, and prone to error.
The web app is built using React for the frontend and Node.js for the backend, offering a user-friendly interface for uploading documents and viewing verification results.
- Signature Extraction: Signature detection and extraction using the YOLOv8 model.
- Signature Verification: A Siamese Network model differentiates between genuine and forged signatures.
- Web-based Interface: Easy-to-use interface for uploading documents and displaying verification results.
- The YOLOv8 model was chosen after comparing it with YOLOv5, with YOLOv8m showing superior performance.
- Metrics tracked during training include box loss, classification loss, precision, recall, and mAP, with YOLOv8m achieving perfect signature detection in testing.
- The model was trained on both a primary dataset and the CEDAR dataset.
- Outperformed state-of-the-art methods on the CEDAR dataset, with strong accuracy and balanced precision-recall results.
- YOLOv8 Performance: Achieved perfect signature extraction on test images.
- Siamese Network Performance:
- Primary Dataset: Accuracy: 74.14%, Precision: 74.36%, Recall: 85.29%.
- CEDAR Dataset: Accuracy: 85.41%, Precision: 92.30%, Recall: 75.00%.
- Precision-recall curves indicate a balanced approach to identifying genuine signatures while minimizing false positives and negatives.
User Interface
Confusion Matrix
| Hyperparameter | Value |
|---|---|
| Learning Rate | 0.0001 |
| Batch Size | 16 |
| Number of Epochs | 20 |
| Optimizer | Adam |
| Loss Function | Binary Cross Entropy |
- Clone the repo
git clone https://github.com/Sujan-Koirala021/Signature-Verification-System- Install necessary packages
go to .../server>
npm install
go to .../client>
npm install
- Starting the application
go to .../server> -> starts at 3000 port
npm start
go to .../client> -> starts at 3001 port
npm start
This project demonstrates a successful implementation of machine learning for offline signature verification. By combining YOLO for signature extraction and a Siamese Neural Network for verification, the system provides a robust solution for detecting forged signatures with real-world applicability.