The overlapping effect and fusion protocols of data augmentation techniques in iris PAD

Abstract Iris Presentation Attack Detection (PAD) algorithms address the vulnerability of iris recognition systems to presentation attacks. With great success deep learning methods in various computer vision fields, neural network-based PAD emerged. However, most networks suffer from overfitting due insufficient data variability. Therefore, we explore impact augmentation techniques on performance and generalizability PAD. We apply several generate variability, such as shift, rotation, brightness. provide in-depth analyses overlapping effect these performance. In addition widely used techniques, also propose an selection protocol based assumption that contribute differently Moreover, two fusion are performed for more comparisons: strategy-level score-level combination. demonstrate experiments fine-tuned models one trained scratch network perform datasets Iris-LivDet-2017 competition designed evaluation. Our experimental results show improve many cases. least overlap-based achieves lower error rates networks. Besides, shift strategy exceeds state-of-the-art (SoTA) Clarkson IIITD-WVU datasets.