Non-reference image quality assessment for biometric presentation attack detection

Non-reference image quality measures are used to distinguish real biometric data from data as used in presentation / sensor spoofing attacks. An experimental study shows that based on a set of 6 such measures, classification of real vs. fake iris, fingerprint, and face data is feasible with an accuracy of 90% on average. However, we have found that the best quality measure (combination) and classification setting highly depends on the target dataset. Thus, we are unable to provide any other recommendation than to optimise the choice of quality measure and classification setting for each specific application setting.