Case study: development of an image-based countermeasure for fake irises
Creator of the world's largest commercial deployment of biometric iris recognition technology, IrisGuard Inc. is maintaining a watchlist of individuals who have been expelled from the UAE, without the right to return. The identity of all visitors to the country is checked and compared to the watchlist, using iris recognition.
In spite of the ban on re-entry, many shady characters attempt to fool the iris recognition system, otherwise known to be very accurate, by presenting a decoy instead of their real irises. One of the most popular spoof attempts is the use of specially crafted contact lenses with patterns printed on them. These patterns are designed to alter the appearance of the real iris underneath.
Answering the call for an efficient countermeasure to contact lens spoofing, we have developed an automatic system that detects printed contact lenses in iris images.
We have created a set of statistical descriptors, which effectively capture the differences between the patterns of a true iris and the patterns printed on a contact lens. Those descriptors, chosen using sophisticated feature selection techniques, have been used to train two probabilistic models of appearance: one of a real human iris, and another one, of an iris covered with a printed contact lens.
We have devised a classification system based on a classifier ensemble which incorporates generative and discriminative dichotomizers. The developed system is able to discriminate between fake and genuine iris images in a split of a second with accuracy exceeding 97% *
If you would like to find out more about this and other state-of-the-art applications of pattern recognition, please contact us.
* LOO error estimate of the Half Total Error Rate in laboratory tests.