Abstract

In this paper, a novel method of writer-independent online signature verification is put forward, employing Siamese-architected recurrent neural networks (RNNs). A bidirectional strategy for accessing past and future contexts, as well as Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU) systems, are investigated. The study compares the benefits and drawbacks of each recurrent Siamese network while doing a thorough examination of system performance and training time. The suggested recurrent Siamese networks outperform the most advanced online signature verification systems, as shown by the experimental findings on the BiosecurID database, which has 11,200 signatures from 400 individuals. Furthermore, the study uses the GPDS Synthetic Signature Database to classify signatures using deep learning, more precisely Convolutional Neural Networks (CNNs) based on the GoogLeNet architecture (Inception-v1 and Inception-v3). The models receive a high level of validation.