Mouse Dynamics as a Behavioral Biometric: Exploration of Novel Visual Representations and Transformer Approaches

Abstract

User authentication based on mouse movements have become increasingly important due to their non-intrusive nature and utility in security and forensic applications. However, existing image-based mouse dynamics representations often fail to adequately capture temporal ordering information and the nuanced characteristics of user mouse behavior, particularly when trajectories overlap due to repetitive user actions. Motivated by these limitations, this thesis investigates deep learning-based approaches for continuous mouse dynamics authentication through novel image representations and transformer-driven architectures. Initially, recurrence plot representations are explored to capture repetitive movement patterns present in mouse trajectories. Subsequently, the recurrence-based framework is extended through the introduction of vicinity plots and recurrence probability maps, which provide a more flexible and informative characterization of mouse behavior.