Sensor Fusion Based Implicit Authentication for Smartphones
Implicit authentication, as a novel identity authentication mechanism, has received widespread attention. However, the performance of implicit authentication still needs to be improved. In this paper, we propose a sensor fusion based implicit authenticati
- PDF / 577,873 Bytes
- 12 Pages / 439.37 x 666.142 pts Page_size
- 54 Downloads / 214 Views
Abstract. Implicit authentication, as a novel identity authentication mechanism, has received widespread attention. However, the performance of implicit authentication still needs to be improved. In this paper, we propose a sensor fusion based implicit authentication system to enhance the protection level of the identity authentication mechanism for smartphone. First, the sensor used to characterize user behavior is determined by analyzing the authentication performance of each sensor. Then, considering the practicability of the system, one-class classification algorithm One-Class Support Vector Machine (OC-SVM) is used to train the authentication model. Finally, the decision function of each sensor is weighted and fused to deliver a result. Based on the actual data set of 7,500 samples collected from 75 participants, the effectiveness of the system is verified in different operating environments and varied passwords. The results show that the proposed system can improve the accuracy of identity authentication effectively.
Keywords: Implicit authentication OC-SVM
1
· Pattern lock · Sensor fusion ·
Introduction
With the widespread usage of smartphones, a large amount of private information is stored on the phone, so a convenient and reliable identity authentication system for smartphones is necessary. Pattern lock is widely used because of its memorability [1]. However, it is vulnerable to acoustic attack [2] and videobased attack [3], etc., which will weaken the security of pattern lock. With the enrichment of smartphone sensors, researchers have attempted to use multiple sensors to record user interaction data during pattern input. The user’s behavior data is formed in daily life, which is unique and difficult to be imitated. Through the authentication analysis of the user’s unlocking behavior characteristics, the method of transparently implementing user identity authentication is called implicit authentication [4]. At present, some scholars have carried out research on implicit authentication [5–10]. Implicit authentication records the user’s behavior information c Springer Nature Singapore Pte Ltd. 2020 Z. Hao et al. (Eds.): CWSN 2020, CCIS 1321, pp. 157–168, 2020. https://doi.org/10.1007/978-981-33-4214-9_12
158
D. Shi and D. Tao
through sensors built into the phone. The collected behavior information is automatically authenticated in the background of the mobile phone. In the existing implicit authentication system, to improve the authentication performance, feature extraction and feature selection are carried out on the original data collected by the sensors, but the performance of implicit identity authentication still needs to be improved. A feature vector is usually thus initiated based on the features extracted from each sensor data and input into the authentication model for identity authentication. However, the features extracted by the sensors often contain some confusing information, which leads to classification errors and weakens the authentication performance. Based on the above background, we
Data Loading...
