Sensor-based implicit authentication through learning user physiological and behavioral characteristics

Nowadays, a large amount of users’privacy data is stored in mobile intelligent devices, and authentication systems have become an important defense to protect that data. The traditional explicit authentication, however, is under significant threat from a great deal of side channel attacks, such as those based on computer vision and touch screen attacks. Users’authentication keys are facing a high risk of leakage, whereas implicit authentication is one of the effective methods to resist various side channel attacks. Nevertheless, existing user implicit authentication approaches still have weaknesses in security and usability, which increase users’authentication burdens and pose potential safety issues. To address this problem, our paper proposes a user implicit authentication approach based on physiological and behavioral characteristics, using the most commonly performed actions in daily life to pre-authenticate users. In our method, multiple sensors are utilized to extract physiological and behavioral characteristics of users, and the sensor data is fused by aligning the extracted characteristic data and using the Kalman filter to reduce its noise. Users’authentication results can be determined by comparing the similarities between their reference and authentication samples using the siamese neural network. The experimental results show that our approach improves the performance of implicit authentication while ensuring its security and usability.