Signature Verification Using Convolutional Neural Network

Abstract

Signatures serve as a prevalent means to authenticate individuals, yet a reliable method for accurately certifying signature authenticity remains a pressing need. The solution proposed in this paper aims to assist individuals in discerning signatures to determine their genuineness or potential forgery. Our system endeavours to streamline the signature verification process by employing Convolutional Neural Networks (CNNs). The model is built upon a pre-trained CNN, specifically the VGG-19 architecture. We assessed the performance of our model using well-established signature datasets, comprising a diverse set of authentic signatures obtained from ICDAR [3], CEDAR and Kaggle. The achieved accuracies were 100%, 88%, and 94.44%, respectively. Our analysis indicates that our proposed model excels in distinguishing signatures, especially when they deviate significantly from authentic signatures.