Some people possess an exceptional ability to recognize faces, even those they have only met once or not encountered for many years.

To explore the secret behind this talent, cognitive scientists from Sydney, Australia, conducted a study on 37 individuals known as "super-recognizers", whose facial recognition skills far outperform ordinary people.

In the experiment, participants were shown face photos one by one on a screen and asked to judge whether each face was new or familiar.

Meanwhile, eye-tracking technology was applied to record where and how long they looked. For comparison, the same test was carried out among 68 people with average facial recognition abilities.

The researchers then reproduced the viewing patterns of both groups and fed the data into nine machine learning models that had been trained for facial recognition tasks. These models were then given the same identification task as human participants.

The results showed that models using the viewing patterns of super-recognizers achieved significantly higher accuracy in face-matching than those using patterns from average viewers.

Further analysis revealed that super-recognizers do not spend more time or effort observing faces; instead, they naturally focus on regions of "higher diagnostic value".

Lead researcher James Dunn stated that super-recognizers do not actually perceive more visual information, but their eyes automatically fixate on the most informative facial features that distinguish one person from another.

He emphasized that this ability is not a learnable skill but an automatic and dynamic process unique to their brains.

This study also indicates that facial recognition depends not only on visual processing in the brain but also on the way people examine a face.

The findings are expected to provide valuable guidance for improving future facial recognition systems.