Eye Movement Dynamics Differ between Encoding and Recognition of Faces.

Joseph M Arizpe,Annie W.-Y Chan,Jack W Tsao,Danielle L Noles

doi:10.3390/vision3010009

Abstract

Facial recognition is widely thought to involve a holistic perceptual process, and optimal recognition performance can be rapidly achieved within two fixations. However, is facial identity encoding likewise holistic and rapid, and how do gaze dynamics during encoding relate to recognition? While having eye movements tracked, participants completed an encoding (“study”) phase and subsequent recognition (“test”) phase, each divided into blocks of one- or five-second stimulus presentation time conditions to distinguish the influences of experimental phase (encoding/recognition) and stimulus presentation time (short/long). Within the first two fixations, several differences between encoding and recognition were evident in the temporal and spatial dynamics of the eye-movements. Most importantly, in behavior, the long study phase presentation time alone caused improved recognition performance (i.e., longer time at recognition did not improve performance), revealing that encoding is not as rapid as recognition, since longer sequences of eye-movements are functionally required to achieve optimal encoding than to achieve optimal recognition. Together, these results are inconsistent with a scan path replay hypothesis. Rather, feature information seems to have been gradually integrated over many fixations during encoding, enabling recognition that could subsequently occur rapidly and holistically within a small number of fixations.

Highlights

Eye movement studies have helped in the investigation of the different visual information sampling mechanisms involved in various cognitive processes concerning facial perception, such as identity recognition [1,2], matching [3,4], emotional expression identification [5,6], and other-race identification [7,8,9,10,11], among others
We found that old/new recognition performance increased for the long compared to the short study phase stimulus presentation time, indicating that, unlike for recognition, two fixations do not suffice for optimal face encoding
Our results reveal that eye movement dynamics differ between encoding and recognition of faces and that longer sequences of eye-movements are functionally necessary to achieve optimal encoding than are necessary to achieve optimal recognition

Summary

Introduction

Eye movement studies have helped in the investigation of the different visual information sampling mechanisms involved in various cognitive processes concerning facial perception, such as identity recognition [1,2], matching [3,4], emotional expression identification [5,6], and other-race identification [7,8,9,10,11], among others. The present study aims to help fill this gap by investigating the difference and relationship between the visual processing mechanisms of facial encoding and recognition. Prior eye movement evidence indicates that two fixations suffice for optimal facial recognition and that initial fixations correspond to an optimal location for facial identification information sampling. The preferred location of the initial fixation tended to land over a featureless location just below the eyes, which, according to a Bayesian ideal observer model, corresponds to a location that is optimal for facial information integration

Results

Discussion

Conclusion