Exploring the Use of IrisCodes for Presentation Attack Detection

Abstract

Iris recognition systems are vulnerable to presentation attacks where an adversary uses artifacts such as cosmetic contact lenses, printed eye images, doll eyes, etc. to undermine the system. In this work, we investigate whether IrisCodes, that are commonly used for iris recognition, can be used for presentation attack detection. IrisCodes are binary phasor features extracted from the geometrically normalized iris, where the annular iris region is transformed to a rectangular entity. We demonstrate that including pupil information in IrisCodes improves presentation attack detection performance. Further, we demonstrate that extracting binary phasor information from the un-normalized iris image-referred to as OcularCode in this paper - can further boost the performance. Experiments involving both printed iris images and cosmetic contact lenses from a benchmark dataset suggest that the proposed methods based on binary phasor codes achieve promising results.