Handwritten Signature Authenticity Verification System April 2024:
Developed a cutting-edge system for authenticating handwritten signatures, aimed at revolutionizing signature verification processes and bolstering security measures.
Demo Video :
Technical stack Used in the Project - 
- Utilized the renowned CEDAR dataset, encompassing signatures from 55 individuals written in Latin script, comprising 24 genuine and 24 forged signatures per person, totaling 2,640 signature samples.
- Engineered two distinct PyTorch models:
- SiasemeNetwork1: Employed Binary Cross Entropy as the loss function, achieving exceptional accuracy rates:
- Train Acc: 98.59%
- Val Acc: 98.08%
- Test Acc: 98.11%
- SiameseNetwork2: Leveraged Contrastive Loss for enhanced model optimization, with satisfactory accuracy metrics:
- Train Acc: 65.07%
- Val Acc: 64.88%
- Test Acc: 65.00%
- SiasemeNetwork1: Employed Binary Cross Entropy as the loss function, achieving exceptional accuracy rates:
- Seamlessly integrated the models into a Streamlit application, providing an intuitive interface for signature authenticity verification, thus ensuring user-friendly interaction and streamlined deployment.
The Github code is here