Biometric recognition through eye movements using a recurrent neural network

Eye movement biometrics have traditionally been tackled by using handcrafted features which lead to complex computation and heavy reliance on experimental design. The authors of this study present a general recurrent neural network framework for biometric recognition through eye movements whereby the dynamic features and temporal dependencies are automatically learned from a short data window extracted from a sequence of raw eye movement signals. The model works in a task-independent manner by using short-term feature vectors combined with using different stimuli in training and testing. The model is trained end-to-end using backpropagation and mini-batch gradient descent. We evaluate our model on a dataset with 32 subjects presented with static images, and the results show that our deep learning model significantly outperforms previous methods. The achieved Rank-1 Identification Rate (Rank-1 IR) for the identification scenario is 96.3% and the Equal Error Rate (EER) for the verification scenario is 0.85%.