Detection of Forged Handwritten Signatures Using Deep Architectures

Detecting forged handwritten signatures is essential for maintaining the security of authentication systems in various fields. This study leverages Convolutional Neural Networks (CNNs) and an upgraded transfer learning model, MobileNet, to enhance the accuracy of signature forgery detection. The research begins with a CNN-based model and later integrates MobileNet for improved performance. The Kaggle "Handwritten Signatures" dataset was utilized, with data augmentation and early stopping applied to optimize training. The results demonstrate that MobileNet significantly outperforms the baseline CNN model, achieving high accuracy and robustness in identifying forged signatures.