Abstract
Earlier studies have examined saccadic eye movements while searching for a single target in noise. Here we investigate saccades while humans actively search for multiple targets in brief displays and subsequently identify all potential target locations. Under these circumstances, saccades need to be efficient to maximize the number of correct decisions.
Methods: The search display had six potential signal locations, 3 degrees from a central fixation spot. Each location had an independent probability of having a target, so a trial could have from 0 to 6 targets. The target was a vertical string of 5 dots presented among randomly positioned noise dots at varying noise levels. Observers actively searched the brief display and then chose all potential target locations with a cursor. The display duration varied from 350 to 1150 ms.
We compared human eye movements to the prediction of a Bayesian model that determined where to look. We assumed that each location had an oriented filter selective for the target. The probability of a target at a location depends on the prior probability of a target, and on the likelihood of a particular filter response given that the location contains a target or not. The filter response tends to be high when a target is present and low when no target is present. Intermediate target responses are equally likely to be target or noise. The model predicts that for brief displays, these uncertain locations are inspected first.
Results & Conclusion: At low and moderate noise levels, observers made few saccades, typically to likely target locations. At high noise levels observers made saccades to uncertain locations as well as to likely target locations. This finding differs from single-target search where saccades are directed to the most likely target locations.
Supported by AFOSR grant FA9550-05-1-0151, and by Smith Kettlewell.