A Low-memory Mobile Phone Continuous Authentication Scheme Based on Multimodal Biometric Information

At present, the existing continuous identity authentication technology using multi-modal data relies on various sensor devices in mobile phones, and effective continuous identity authentication can be realized by extracting complex features from these data. Due to the low robustness of the single-mode biometric model, it can not be effectively authenticated. In order to achieve better verification effect, a variety of sensor data are used, which also leads to large memory consumption, while a simple classifier leads to poor accuracy and high calculation cost. In order to solve these problems, we propose a low memory consumption continuous authentication scheme based on motion state and touch behavior. Aiming at the problem of memory consumption in current continuous identity authentication, a new memory processing scheme is proposed to process sensor data, and a new data set of time expansion behavior is constructed, which achieves the goal of greatly reducing the memory consumption of mobile devices without reducing user identity information. Using the processed data, this paper establishes a new classification model, and designs a network based on CNN-LSTM-Attention algorithm as an authentication algorithm, which can better analyze, learn and classify the user's touch behavior, and is superior to the traditional classification model. In the evaluation experiment of 92 test objects, the precision is 98.97% and the F1 value is 99.42%.