Super-recognizers sample visual information of superior computational value for facial recognition

Area of Science:

• Cognitive Neuroscience
• Visual Perception

Background:

• Super-recognizers possess exceptional facial recognition skills, a prime example of biological visual expertise.
• Previous research suggests exploratory viewing during learning drives this expertise, but its functional impact on face identity processing is debated.

Purpose of the Study:

• To quantify the computational value of facial information samples.
• To investigate the utility of information sampling strategies employed by super-recognizers for face identity processing.

Main Methods:

• Developed a novel method to quantify the computational value of facial information samples.
• Reconstructed retinal information acquired during face learning using eye-gaze data.
• Evaluated information utility using nine deep neural networks (DNNs) for face identity matching.

Main Results:

• Identity matching accuracy improved in all DNNs when using visual information sampled by super-recognizers versus typical viewers.
• The advantage was not solely due to the quantity of information sampled.
• Both quantity and quality of retinal information contribute to individual differences in face processing.

Conclusions:

• Active visual exploration and attentional mechanisms are crucial for developing visual expertise in face recognition.
• Super-recognizers' superior face processing stems from both the amount and quality of information they encode.
• Eye-gaze behavior during learning plays a functional role in enhancing face identity processing.