Biometric keystroke barcoding: A next-gen authentication framework. (1st September 2021)

• Record Type: Journal Article
• Authors: Alpar, Orcan
• Is Part Of: Expert systems with applications. Volume 177(2021)
• Publication Date: 2021-09-01
• Subjects: Keystroke authentication -- Biometrics -- Complex wavelets -- Scalogram -- Barcode -- CWT -- SVM; Expert systems (Computer science) -- Periodicals; Systèmes experts (Informatique) -- Périodiques; Electronic journals; 006.33
• DOI: 10.1016/j.eswa.2021.114980
• Languages: English
• ISSNs: 0957-4174

Abstract

Highlights: We introduced the keystroke barcodes for the first time in the literature. The barcodes carry information on the unique biometric password entering styles. The keystroke data is quantized for frequency vs time localization using CWT. The complex Frequency B-spline wavelets are investigated for scalogram conversion. We achieved 1.834% EER for 109 real and fraud attempts with a training size of 4. Abstract: Investigation of new intelligent solutions for user identification and authentication is and will be essential for enhancing the security of the alphanumeric passwords entered on touchscreen and traditional keyboards. Extraction of the keystrokes has been very beneficial given the intelligent authentication protocols operating in time-domain; while the time-domain solutions drastically lose their efficiency over time due to converging inter-key times. Realistically reflecting the habitual traits, the frequency-domain solutions, however, reveal unique biometric characteristics better, without any risk of convergence. On the contrary, the existing frequency-based frameworks don't provide storable biometric data for further classification of the attempts. Therefore, we propose a novel barcoding framework converting habitual biometric information into storable barcodes as very low-size barcode images. The key-press times are extracted and turned into pseudo-signals exhibiting binary-train characteristics for continuous wavelet transformation (CWT). The transformed signals are primarily categorized with 4-scale scalograms by various complex frequency B-spline wavelets and subsequently superposed to create the unique barcodes. One-class support vector machines (SVM) is employed as the main classifier for training and testing the barcodes and very promising results are achieved given the lowest equal error rate (EER) of 1.83%.