New user authentication method based on eye-writing patterns identified from electrooculography for virtual reality applications

• Published in: Biomedical engineering letters Vol. 15; no. 1; pp. 95 - 104
• Main Authors: Kim, HyunSub, Kim, Chunghwan, Kim, Chaeyoon, Kwak, HwyKuen, Im, Chang-Hwan
• Format: Journal Article
• Language: English
• Published: Korea The Korean Society of Medical and Biological Engineering 01.01.2025; Springer Nature B.V
• Subjects: Authentication; Biological and Medical Physics; Biomedical Engineering and Bioengineering; Biomedicine; Biophysics; Computer applications; Data acquisition; Electrooculography; Engineering; Medical and Radiation Physics; Original Article; Passwords; Similarity; Virtual reality; Electrooculography; Biometric system; User authentication; Virtual reality
• ISSN: 2093-9868; 2093-985X; 2093-985X
• DOI: 10.1007/s13534-024-00426-8

Demand for user authentication in virtual reality (VR) applications is increasing such as in-app payments, password manager, and access to private data. Traditionally, hand controllers have been widely used for the user authentication in VR environment, with which the users can typewrite a password or draw a pre-registered pattern; however, the conventional approaches are generally inconvenient and time-consuming. In this study, we proposed a new user authentication method based on eye-writing patterns identified using electrooculogram (EOG) recorded from four locations around the eyes in contact with the face-pad of a VR headset. EOG data acquired during eye-writing a specific pattern are converted into a ten-dimensional vector, named a similarity vector, by calculating similarity values between the EOG data for the current pattern and ten pre-defined template patterns using dynamic positional warping. If the specific pattern corresponds to password, the similarity vector will have shorter distance to a similarity vector of the pre-registered password than an individually pre-determined threshold value. Nineteen participants were instructed to eye-write ten template patterns and five designated patterns to evaluate the performance of the proposed method. A specific user’s similarity vectors were computed using the other users’ template EOG data, employing the leave-one-subject-out cross-validation scheme. The proposed method exhibited an average accuracy of 97.74%, with a false accept rate of 1.31% and a false reject rate of 3.50%. The proposed method would provide a new effective way to secure private data in practical VR applications with edge devices because it does not require heavy computational burden.