Behavioral Biometrics for Human Identity Corroboration based on Gesture-Signature with Deep Learning

Identity corroboration has gained a substantial deal of attention with the high usage of smart devices and dedicated systems accommodating sensitive data and applications magnitudes. In this paper, we develop a behavioral biometric authentication system based on deep learning. Specifically, an android application is developed as a dedicated tool for capturing the touch behavioral biometrics information, the electronic signature information alongside their corresponding accelerometer and gyroscope sensor readings. Finally, the generated sensor readings are fed into a multi-input convolutional neural network architecture for classification. Extensive experimental results show that our proposed approach uniquely identifies users with a classification performance of 93.46% as compared to other baseline approaches, where only a single sensor reading is considered.