Electromyograph and keystroke dynamics for spoof-resistant biometric authentication

Biometrics has come a long way over the past decade in terms of technologies and devices that are used to verify user identities. Three of the more well studied modalities in this field are the face, iris and fingerprint, with the latter two reporting very high user identification/verification rates. In the biometric community there has been little work in studying biomedical signals for user recognition purposes. In this paper, we propose using electromyograph (EMG) signals as a person's biometric signature. The EMG records the motor unit action potentials (MUAP) during any physical motion. Our study is done within the context of a person using a keyboard to type a password or any other fixed phrase. Along with EMG signals, we log key press times for the user and study the feasibility of using this data too as a biometric feature. Keypress timings alone if used as a biometric, are very easy to spoof and hence we fuse this modality with EMG signals. In order to classify these features, we use subspace modeling as well as Bayesian classifiers. The experiments have been performed within the context of a user typing a fixed pass phrase at a workstation. The idea is to monitor both biometric modalities when this action is performed and study user verification across data capture sessions and within capture sessions. Our approach yields high values of verification rates, which shows the promise of using these modalities as user specific biometric signatures.