User Keystroke Authentication Based on Convolutional Neural Network

Abstract

Biometric authentication technology has become an important measure to protect user information security. Among them, keystroke authentication technology has attracted the attention of many researchers because of its low cost and high benefit. In recent years, various methods such as statistical methods and integrated models have been increasingly used in user keystroke authentication and have achieved relatively good results. However, few people try to convert keystroke data into images and tap spatial information between keystroke data. In this paper, we used a new way to convert keystroke data into images, then we use a binary model based on a Convolutional Neural Network (CNN) for each genuine user and try to import the transformed images into the CNN model. In this way, we can dig out the “spatial information” of keystroke features, which is the advantage over other models, and this method is first proposed for keystroke behavior authentication. In addition, we have also tried data augmentation, relu activation functions and dropout methods to reduce overfitting. In the end, we got an accuracy of 96.8%, which is about 10% higher than the previous work.

Supported in part by the National Basic Research Program of China (973) under Grant No. 2013CB329102, and in part by the Natural Science Foundation of China (NSFC) under Grant No. 61003283.