An advanced AI-based Signature Verification System using Siamese Neural Networks to detect and verify handwritten signatures. With a clean Gradio web interface, users can simply upload two signatures to find out if they match — perfect for banking, legal, and security use cases.
This project applies deep learning to compare two handwritten signatures and determine if they belong to the same person. It uses a Siamese CNN model trained to learn feature similarity between genuine and forged signatures.
The app provides:
best_signature_model.h5)| Component | Technology |
|---|---|
| Deep Learning | TensorFlow, Keras |
| Model | Siamese CNN |
| Web UI | Gradio |
| Image Processing | OpenCV, Pillow (PIL) |
| Visualization | Matplotlib |
| Data Handling | NumPy, Scikit-learn |
| Language | Python 3.7+ |
signature_verification.py to your Colab session!pip install gradio tensorflow opencv-python-headless pillow numpy matplotlib scikit-learn
%run signature_verification.py
A public link will appear — click to open the app in your browser.
git clone https://github.com/shahram8708/AI-Based-Signature-Verification-System.git cd AI-Based-Signature-Verification-System python -m venv signature_env # For Windows: signature_env\Scripts\activate # For Mac/Linux: source signature_env/bin/activate pip install -r requirements.txt python signature_verification.py
Open browser at: http://localhost:7860
AI-Based-Signature-Verification-System/
├── signature_verification.py # Main application file
├── best_signature_model.h5 # Trained model (auto-generated after training)
├── requirements.txt # Dependencies
├── demo.PNG # Demo screenshot
├── README.md # Documentation
└── signature_data/ # Signature dataset directory
├── genuine/ # Real signatures
└── forged/ # Fake signatures
| Parameter | Value |
|---|---|
| Architecture | Siamese CNN |
| Input Size | 150x150 grayscale |
| Feature Dim | 128D Embeddings |
| Loss Function | Binary Crossentropy |
| Optimizer | Adam (lr=0.001) |
| Training Epochs | Up to 30 (Early Stopping) |
| Augmentation | Rotation, Scaling |
This project is licensed under the MIT License.
You are free to use, modify, and distribute it with credit to the author.
Shah Ram — AI/ML Engineer & Deep Learning Enthusiast
Connect with me on GitHub
"The best way to predict the future is to create it." – Peter Drucker
Built with ❤️ using TensorFlow & Gradio