NiohSign: A Siamese Neural NetworkApproach for Signature Authentication
Anthony Effiok · NORMA · 2020
Signatures continuously play important roles as an accurate means of identification and also as a means to verify permissions given for tasks. The process of verifying if an appended signature for a task is forged or genuine plays a largely important role when trying to prevent acts relating to fraud or worse impersonation. Objective- This project aims to create a network that is capable of identifying if a signature is forged or authentic through the use of Siamese neural networks which is essentially a twin neural network. Methodology- The network to be used a s the base network is a model of the ResNet network known as InceptionResNetV2 model. The dataset used is a signature dataset that contains the signatures of 260 individuals with 100 in Bengali and 160 in Hindi. Results- The Model is tested on the signatures appended in Hindi and is seen to perform well with an accuracy of 81.71%. Keywords - Signature authentication, Siamese neural network, Artificial neural network