Online Signature Profiling using Generative Adversarial Networks

A signature is an ability learned by humans from an elementary age. The skill to generate one's own exclusive signature along with imitating another writer's signature is a challenging and complex task. In real time scenarios like E-Commerce and M-Commerce payments, user verification based on online signatures constrain the verification framework needs to be trained extensively with huge samples, which unfeasible to obtain. Hence, as a solution, in this paper, we propose a first its of kind of attempt in which an intelligent framework tries to learn the online signatures of a writer using Deep Generative Adversarial Networks (DGANs). Thorough experimental analysis on three widely used datasets MCYT-100, SVC, SUSIG confirms the supremacy of the method and boost confidence in real time deployment of our framework in data centric applications like offline signature verification, forged document detection, etc.