Handwritten Signature Detection using Deep Learning Approach

Abstract

Signature verification is a type of biometric technology that is widely used for personal identification. In many commercial circumstances, such as the payment of a bank check, the signature verification procedure is based on the human analysis of a single known sample. For verification purposes, the majority of organizations primarily focus on the signature's visual appearance. The signing of a signature is required for many documents, including forms, contracts, bank checks, and credit card transactions. Subsequently, it is of the highest significance to have the option to perceive marks precisely, easily, and as soon as possible. A manually written signature is normally rehearsed course for affirming the legitimacy of authoritative records. Because the signature varies on a regular basis and may alter in response to factors such as age, behavior, and the environment, its verification is crucial. To verify the signature, a deep learning model based on the CNN architecture is presented in this paper. We have data that has been gathered from volunteers who have consented to provide it. About 4200 photos, separated into 21 classifications, were used in this research. For each participant, we worked with a distinct class. After them, classes are named. 200 data are contained in each class. The dataset is classified using five different classification methods: CNN, VGG 16, VGG 19, Inception v3, and MobileNet v2. Here is separated the dataset into three sections for training, testing, and validation: 80%, 10%, and 10%, respectively. We have used multiple deep learning algorithms like Convolutional Neural Network it with image processing tools to form a better structure, leading to higher accuracy of 99.41% in VGG19.