Improving Ocular Recognition with Synthetic Data

Ocular recognition is a form of biometric identification that gained significance after the widespread use of face masks during the COVID-19 pandemic. Modern recognition systems are built upon deep learning methods, such as Convolutional Neural Networks, which require large datasets to be trained. However, acquiring extensive datasets of face and ocular images is hard to obtain due to privacy regulations, raising the question of whether synthetic data can be an alternative or complement. Tests were carried out on recognition models trained on real and synthetic data, with various Python programs. While metrics (e.g. Rank-1 rate) were promising, the overall performance, especially Equal Error Rate, did not meet required safety thresholds for a biometric recognition system.