Mobile behavioral biometrics for passive authentication

• We present a first benchmark of HuMIdb, publicly available HCI database. evaluate touchscreen and background sensor data for mobile passive authentication. train recurrent neural network with triplet loss each individual modality. A multimodal system is achieved through the fusion modalities at score level. Our results show suitability behavioral biometrics Current user authentication systems based on PIN codes, fingerprint, face recognition have several shortcomings. Such limitations been addressed in literature by exploring feasibility devices biometrics. In this line research, work carries out comparative analysis unimodal biometric traits acquired while subjects perform different activities phone such as typing, scrolling, drawing number, tapping screen, considering simultaneous (accelerometer, gravity sensor, gyroscope, linear accelerometer, magnetometer). experiments are performed over 1 https://github.com/BiDAlab/HuMIdb one largest most comprehensive freely interaction databases to date. separate Recurrent Neural Network (RNN) implemented single Then, weighted carried our experiments, discriminative magnetometer, whereas among touch tasks best keystroke fixed-text scenario. all cases, very beneficial, leading Equal Error Rates (EER) ranging from 4% 9% depending modality combination 3-second interval.