Deep Learning Leads to Death-proof Liveness Detection for Iris Biometrics

“They trained their deep learning system on a database of 574 near-infrared post-mortem images, as well as a database of 256 images of irises from live subjects…”

Polish researchers are working on technology that can prevent fraudsters from using the iris biometrics of dead people.

The researchers, Mateusz Trokielewicz, Adam Czajka, and Piotr Maciejewicz, have published a new paper outlining their technique. Using deep learning technology, the researchers say they have devised a system that can differentiate between a living or dead eye in nearly 99 percent of trials, with the accuracy of their system increasing with more time after death.

They trained their deep learning system on a database of 574 near-infrared post-mortem images, as well as a database of 256 images of irises from live subjects; and they say their research paper is the first to delve into presentation attack detection concerning post-mortem spoofing.

It’s a niche area of study, to be sure, but it’s one with increasing applicability as the use of iris recognition technology becomes more widespread. Samsung in particular has pioneered iris-based authentication on its latest smartphones, and it’s plausible that at some point criminals could, for example, try to use the iris biometrics of dead individuals to make purchases on a platform like Samsung Pay, which uses iris scanning for authentication.

In any case, the post-mortem spoofing research could help to advance the field of iris biometrics more broadly with its insights into the differences between living and dead subjects.

Source: Cornell University Library

(Originally posted on FindBiometrics)