Augmented Computer User Login Authentication Using Classifying Regions of Keystroke Density Neural Network

摘要

We present an authentication system using classifying regions of keystroke density based on a neural network architecture with two types of connections: (1) weight vector W and (2) dispersion vector V. In the learning phase, the weight vector W adapts to users' keystroke exemplars, and dispersion vector V adapts to dispersion in the users' keystrokes. Here W represents users' keystroke pattern, and V represents the radius for the regions of density of users' keystrokes. The system consists of three phases: (1) training, (2) validation, and (3) testing. The system learns W and V during training, and adjustment of parameters SF and PS (see Section 3) is done during validation. During testing, classification results in strengthening the vector W, thereby adapting to changing users' typing pattern. We achieved up to individual 0% IPR and 0% FAR. Our highest results are 1.36% IPR and 2.31% FAR. These results compare favorably to the results reported in the literature.