Optimized SVM Parameters with Googlenet Model for Handwritten Signature Verification

Abstract

As far as behavioral biometric authentication procedures go, signature verification is at the top of the list. The most popular form of user authentication, a signature is like a "seal of approval" that confirms the user's permission. The primary objective of this verification technique is to distinguish between real signatures and those that have been forgeried by imposters. In order to learn characteristics from the pre-processed real and fake signatures, Convolutional Neural Networks (CNN) were used in this research. The model utilized to train the CNN was the Inception V1 architecture (GoogleNet). To make the network larger rather than deeper, the design makes advantage of the idea of having various filters on the same level. A small number of publicly accessible datasets, including CEDAR, the BHSig260 signature corpus, as well as UTSig, are used to evaluate the suggested approach in this study.