Evaluating the Sensitivity of Face Presentation Attack Detection Techniques to Images of Varying Resolutions
In the last decades, emerging techniques for face Presentation Attack Detection (PAD) have reported a remarkable performance to detect attack presentations whose attack type and capture conditions are known a priori. However, the generalisation capability of PAD approaches shows a considerable deterioration to detect unknown attacks. In order to tackle those generalisation issues, several PAD techniques have focused on the detection of homogeneous features from known attacks to detect unknown Presentation Attack Instruments without taking into account how some intrinsic image properties such as the image resolution or biometric quality could impact their detection performance. In this work, we carry out a thorough analysis of the sensitivity of several texture descriptors which shows how the use of images with varying resolutions for training leads to a high decrease on the attack detection performance.