Offline-Signature Verification System using Transfer Learning VGG-19

Abstract

Nowadays, Signature verification is one of the most common and effective biometric systems that used to recognize people in many institutions. In modern era of technology, advanced neural networks have provided us an option to solve this issue. In this study, The Robinreni Signature Dataset was utilized to classify the signatures of 64 people, each of whom had 64 original signatures and 64 fake signatures. One of the most popular CNN architecture, namely, VGG19, were used. Firstly, the dataset was distributed accordingly 1649 and 500 for training and validation. Secondly, preprocess the data to train the model. After that the model training process is started using transfer learning approach. Obtained experimental results that VGG19 is best suited for datasets with a validation accuracy of 98.79%.. Everyone has their own unique signature that used to identify and verify important documents and legal transactions. Our study shows the effectiveness of VGG19 for Signature Verification task. The findings will aid in the development of more effective Deep Learning-based signature verification methods.