Deep Learning Based Dynamic Signature Analysis for Multiclass Classification

In the current era of digital authentication, the demand for robust identity verification methods is ever increasing. This paper explores advanced techniques for multiclass signature recognition within the realm of deep learning. The study addresses the intricate challenge of distinguishing diverse signatures across a spectrum of users with machine learning as well as deep learning techniques such as random forest classifiers, MLP, bidirectional GRUs. These methods excel in capturing spatial and temporal patterns within signature data, demonstrating their prowess in recognition. Performance evaluation is conducted using comprehensive metrics like EER, EAR, AUC and ROC curves. These metrics provide insights into the model’s capacity for classification accuracy and discriminative power in multiclass signature recognition. The experiments have been performed on a diverse dataset of 22 users, with each signature exhibiting 144 distinct features such as (x, y) coordinates of position of pen, pen pressure and pen tilt angles. An in-depth analysis of AUC and ROC results for each class showcases the model’s proficiency in user and class differentiation.