Samir Brahim Belhaouari

Abstract. This paper details the IJCB 2022 Mobile Behavioral Biometrics Competition (MobileB2C), which benchmarks mobile user authentication systems using behavioral biometric traits acquired during typical Human-Computer Interaction (HCI). The competition utilized the BehavePassDB database and involved four tasks: keystroke, text reading, gallery swiping, and tapping, incorporating data from touchscreen and background sensors. The HBKU CS Lab Team employed a novel approach using Discrete Wavelet Transform (DWT) to deconstruct signals into wavelet-basis functions. Let the wavelet transform be denoted as \(D[a, b] = \frac{1}{\sqrt{b}} \sum_{m=0}^{p-1} f[tm] \phi \left(\frac{tm - a}{b}\right)\) where (a) and (b) are integers representing translation and compression, respectively. Post DWT, the data are recursively averaged and transformed into image representations, which are input into a siamese neural network with contrastive loss, represented by \(c_{A_j}^i, c_{B_j}^i = W(D_j^i)\). This method demonstrated superior performance in gallery swiping and tapping tasks with AUC scores of 61.54% and 59.58%, respectively, confirming the efficacy of the proposed biometric authentication system.
    

Illustration of the proposed experimental design.