Abstract
Introduction: Previously, we have shown that eye fixation patterns during a quick, difficult facial identification task are highly observer-specific, with these differences mirroring idiosyncratic fixation-dependent task ability (Peterson, 2009). Here, we extended this exploration to the task of emotion recognition. Specifically, we investigated the optimality with which humans adapt their eye movements to changing task demands during face recognition. Methods: We implemented an ideal observer limited by a human-like foveated visual system in order to evaluate the expected performance for each possible fixation location. In order to assess human strategy optimization we ran observers in two tasks. In both, observers began each trial by fixating along the edge of a monitor. A face embedded in white noise would then appear in the middle of the screen for 350ms during which observers made a single eye movement. In one task, observers were shown and then asked to identify one of ten faces. In the second task, observers were shown either a smiling face or a neutral face and asked to choose the displayed emotion. Results: Ideal observer results show that a shift in eye movement strategy between identification and emotion recognition downward toward the mouth is optimal. This move is driven largely by the differences in the locations of information concentration between the two stimuli and task types. Humans continued to show individualized fixation patterns across both tasks. Furthermore, humans showed a significant shift in gaze locus between the two conditions on a subject-by-subject and group basis. However, fixations did not shift as much as optimality would suggest. A foveated system with differential upper-field and lower-field visibility (Cameron, 2002) is able to explain the pattern of eye movements. Conclusion: Humans optimally adapt their eye movements depending on the face recognition task at hand and on the individual's observer-specific, fixation-dependent ability.
NIH-EY-015925, NSF-DGE-0221713.