Synthetic eye images for pupil dilation mitigation

It is known that the probability of a false non-match increases with difference in pupil size between the enrolled eye image and the probe eye image. To address this issue, recent research showed that enrolling multiple images for an eye, with the images selected to have different dilation values, improves performance. However, it may not always be practical to acquire iris images for a range of different dilation values at enrollment time. Therefore, we consider a process of taking a single real image and synthetically dilating or constricting the pupil to a range of different dilation values. With this approach, multiple enrollment samples can be generated from one original real image. Our results show that there is a marked improvement in recognition accuracy by enrolling a set of synthetic eye images to cover a range of dilation values.