Online Signature Verification System Using Convolution Neural Network and DL

Abstract

Signature verification is an important biometric authentication technique that is widely applied in identification verification, banking and legal systems. Traditional methods often depend on manual inspection or rule-based algorithms, which suffer from impurities and disabilities. The project proposes a firm nervous network (CNN)-based signature verification system to accurately differentiate between real and forged signatures using deep learning techniques. The system contains major stages including data collection, preprosying, CNN-based feature extraction and classification. It is evaluated by using publicly available datasets such as cedar, GPD and MCYT -75. Performance is measured using matrix such as accuracy, accurate, recall, similar error rate (EER), and ROC curve analysis. Experimental results indicate that CNN models effectively learn signature patterns and suit the real-world variations, which improve traditional techniques such as SVMs, HMMs and handicraft facilities methods. While the system improves accuracy and automation, there are challenges such as computational complexity, efficient forgery and intra-class variability. Future improvement aims to increase safety and scalability through blockchain integration, online verification, multimodal biometric fusion and hybrid deep learning models. Overall, the proposed system provides a sharp, accurate and reliable solution to detect safe certification and fraud in various fields.