A New Approach for Biometric Identification Based on Eye Movement Signal

A new approach for eye-movement-based biometric identification was proposed in the presented research. It was based on three features extracted from horizontal eye movement signals. They were percentage changes in point-to-point velocity, acceleration, and jerk. Such a feature set was used to feed a simple Long Short–Term Memory network. Two datasets were utilized using the same stimulus paradigm – jumping point. The datasets differed in the number of stimulus positions (27 and 50), number of sessions (2 and 9), and number of subjects taking part in the experiments (21 and 14). The elaborated solution turned out to be effective in subject recognition at the 75% – 93% level, depending on the dataset and number of samples used for model training and testing. The investigation of the influence of the session number on the model’s performance was also conducted. The number of sessions was shown, below which making reliable identification is difficult according to the proposed approach.