What draws the eye? Since the human visual
system is foveated, there is something of an chicken-and-egg-problem concerning
how we decide where to look when we shift our gaze about the world. We don’t
want to waste energy looking at uninteresting or uninformative things, yet how
do we know what would be useful to look at before we look at it? To answer this
question, we have been using a modified version of the discrimination image
paradigm of Beard and Ahumada (1998) combined with accurate eye-movement monitoring
during visual search. The animated figure below illustrates the fundamental
paradigm. A subject searches for a target embedded in 1/f noise, in this case a
64-pixel-square black-over-white edge. The noise about each fixation point is
sampled and averaged to form a discrimination image which can be thought of as
the template for which the subject is actually searching. The panel on the top
shows a simulated pattern of fixations (a “scanpath”) from a single trial. The
two panels on the bottom show an actual evolving discrimination image in which
each update represents the addition of ~100 trials. The bottom left panel shows
the simple average and the lower right figure shows pixels that are significantly
above or below the mean at a 95% confidence level.
