Iris recognition has developed rapidly to a mainstream field with active researchers in a wide variety of fields. What makes this particular biometric attractive is because the iris’s rich texture offers a strong biometric cue for recognizing individuals. With this in mind, iris recognition systems typically consist of the following modules leading to an identification determination.
- Acquisition – This module obtains a two-dimensional eye image using a monochromatic charge coupled device (CCD) camera that is sensitive to the near infrared (NIR) spectrum.
- Segmentation – This module isolates the information corresponding to the iris texture from other ocular structures in its vicinity such as the pupil, the sclera, eyelids, and eyelashes.
- Normalization – This module invokes a pseudo-polar transformation to transform the annular region of the iris to an equivalent rectangular representation describing the iris texture.
- Encoding – This module uses a feature extraction routine (such as Gabor filtering) to produce binary code.
- Matching – This module determines how closely the produced code matches the encoded features already stored in an iris recognition database.
Although this technology performs well on large datasets, it still has its challenges including understanding and mitigating the effects of pupil motion on iris identification performance. It has been shown that variations of these physiological effects negatively impact iris matching performance. Therefore, there is a need to find ways to mitigate these biological effects as well as holistically understand their impact on matching performance.
At CRC, we examine the reliability of these algorithms from both acquisition and algorithmic perspectives. From an acquisition perspective, we actively assess the overall quality of iris images and videos including considering different physiological variations such as pupil dilation, iris texture changes, and ocular gaze variations. From an algorithmic perspective, we characterize the impact of these quality variations where we model and assess the statistical changes in their genuine and impostor distributions as well as in their overall iris recognition performance.