Abstract
Traditional identity authentication using IDs and passwords cannot cope with the problem when IDs and asswords are leaked or when spoofing occurs. To solve this problem, continuous authentication using behavioral features, which is a biometric authentication method, is effective. In this paper, we use OpenPose to extract features from the skeletal coordinates of the fingers during the Enter keystroke, which is a frequent input keystroke. The purpose of this paper is to evaluate whether behavioral and physical features are included in the hand geometry during Enter keystrokes, and to discuss the data obtained from the experiments conducted in this paper. Keyboard keystroke video data sets were obtained from 20 participants with different levels of experience in keyboard operation and were evaluated using the equivalent error rate. The results showed that the mean equivalent error rate after removing outliers was 5.6%, suggesting that this feature is effective for continuous authentication.
