This project aims to automatically detect whether a signature is genuine or forged using deep learning. Trained on the CEDAR dataset, our CNN-based model analyzes signature patterns with high accuracy
The Sigma uses a Convolutional Neural Network (CNN) trained on the CEDAR dataset to classify signatures as genuine or forged. The model takes grayscale signature images as input, processes them through multiple convolutional and pooling layers to extract spatial and texture-based features, and finally classifies them using dense layers and a sigmoid output.
To improve accuracy and robustness, the training process includes data augmentation techniques such as rotation, scaling, and shifting. Additionally, a document preprocessing module is integrated to detect and crop the signature region from full documents before classification. The system achieves high validation accuracy and performs well even with real-world document inputs.
During the Hackathon, we transformed our idea into Reality. Here's what we accomplished during this time: 1)Collected and preprocessed CEDAR dataset (with data augmentation) 2)Built and trained a CNN model for binary classification (genuine vs forged) 3)Achieved over 98.8% validation accuracy 4)Developed a working Streamlit-based web app for real-time verification
N/A