Effects of Compression on Attention-Based IRIS Presentation Attack Detection

Biometric recognition systems can automatically recognize individuals using their physical or behavioral characteristics, and they are thus adopted in many applications requiring strong authentication mechanisms. The iris plays a pivotal role among the traits employed for biometric purposes, mainly thanks to the high recognition performance achievable with this modality. While techniques further improving current recognition capabilities are being developed, ensuring the security of these systems against attacks is becoming an increasingly pressing need. This paper deals with presentation attack detection (PAD) for iris recognition, analyzing the effects of image compression on the effectiveness of data-driven approaches. The conducted tests rely on attention-based frameworks, namely vision transformers, to perform PAD while providing suitable tools to argue on the decisions' explainability. The obtained results demonstrate the effectiveness of the employed transformer-based PAD, and the influence of compression on the achievable error rates.