Predicting Iris Vulnerability to Direct Attacks Based on Quality Related Features
A new vulnerability prediction scheme for direct attacks to iris recognition systems is presented. The objective of the novel technique, based on a 22 quality related parameterization, is to discriminate beforehand between real samples which are easy to spoof and those more resistant to this type of threat. The system is tested on a database comprising over 1,600 real and fake iris images proving to have a high discriminative power reaching an overall rate of 84% correctly classified real samples for the dataset considered. Furthermore, the detection method presented has the added advantage of needing just one iris image (the same used for verification) to decide its degree of robustness against spoofing attacks.