SECURE SIGNATURE VERIFICATION: FRAMEWORK DEVELOPMENT AND PRACTICAL IMPLEMENTATION

This project presents the design and implementation of an offline signature verification system using MobileNetV2, a lightweight convolutional neural network model from TensorFlow Hub. The system aims to enhance the security and reliability of biometric authentication by automatically distinguishing between genuine and forged handwritten signatures. The model leverages the efficiency of MobileNetV2 to extract high-level feature vectors from preprocessed grayscale signature images. These features are compared against reference signatures using similarity metrics such as cosine distance to verify authenticity. The system architecture comprises a Flask-based web application for signature upload and result display, with backend modules handling image preprocessing, model inference, and result computation. Preprocessing techniques such as image resizing, normalization, and noise reduction are employed to ensure uniform input formatting. The model was fine-tuned on a combination of public and custom signature datasets and evaluated using metrics including accuracy, precision, recall, and F1-score.