This work demonstrates the first fully operational iris-recognition system for 4-6 week old infants. Its novel contribution is a co-designed pipeline: a purpose-built near-infrared iris sensor sized for infants, a new segmentation model that reliably isolates iris texture in infant eyes (where adult-tuned segmenters fail), texture-encoding matchers on top of it, and a StyleGAN generator that synthesizes privacy-safe artificial infant iris images to seed further research. Evaluated on NIR images from 17 infants, the system achieved a 3% Equal Error Rate and 99% AUC, far surpassing adult state-of-the-art methods applied to the same infants (EER>=20%, AUC<=88%), evidencing that infant irises carry extractable, stable biometric features.
This computer-vision paper studies the feasibility of iris recognition as a biometric identifier for 4-6 week old infants, motivated by preventing newborn swapping and abduction and improving post-na…