Overview
- The peer‑reviewed study, published Nov. 5 in Proceedings of the Royal Society B, finds that super‑recognizers target facial regions with more diagnostic identity information.
- Researchers used eye tracking from 37 super‑recognizers and 68 typical observers to rebuild retinal input and evaluate it with nine deep neural networks trained for face recognition.
- Across all models, AI achieved higher matching accuracy when fed visual samples taken by super‑recognizers, indicating a consistent computational advantage.
- The advantage persisted when the total amount of viewed information was equated, pointing to quality of sampled regions rather than quantity or viewing time.
- Authors and outside experts note the work used static, controlled images, leaving open whether the strategy generalizes to dynamic real‑world settings or can be effectively trained in typical observers; the UNSW Face Test is available to the public.