Modelling Primate Visual Attention - Computational Neuroscience: A Comprehensive Approach - page 602

lated, as revealed by psychophysical [19, 28, 47, 48], physiological [1, 6, 29, 45], and
imaging [7, 37] studies. Directing covert attention thus involves a number of sub-
cortical structures that are also instrumental in producing directed eye movements.
These include the deeper parts of the superior colliculus; parts of the pulvinar; the
frontal eye fields in the macaque and its homologue in humans; the precentral gyrus;
and areas in the intraparietal sulcus in the macaque and around the intraparietal and
postcentral sulci and adjacent gyri in humans.

19.3 Bottom-up control

One important mode of operation of attention is largely unconscious and driven by
the specific attributes of the stimuli present in our visual environment. This so-called
bottom-up control of visual attention can easily be studied using simple visual search
tasks as described below. Based on these experimental results, several computational
theories and models have been developed for how attention may be attracted towards
a particular object in the scene rather than another.

19.3.1 Visual search and pop-out

One of the most effective demonstrations of bottom-up attentional guidance uses
simple visual search experiments, in which an odd target stimulus to be located by
the observer is embedded within an array of distracting visual stimuli [52]. Origi-
nally, these experiments suggested a dichotomy between situations where the target
stimulus would visually pop-out from the array and be found immediately, and situ-
ations where extensive scanning and inspection of the various stimuli in the display
was necessary before the target stimulus could be located ( Figure 19.2) . The pop-out
cases suggest that the target can be effortlessly located by relying on preattentive vi-
sual processing over the entire visual scene. In contrast, the conjunctive search cases
suggest that attending to the target is a necessary precondition to being able to iden-
tify it as being the unique target, thus requiring that the search array be extensively
scanned until the target becomes the object of attentional selection.
Further experimentation has revealed that the original dichotomy between the fast,
parallel search observed with pop-out displays and slower, serial search observed
with conjunctive displays represent the two extremes of a continuum of search diffi-
culty [60]. Nevertheless, these experiments clearly demonstrate that if a target differs
significantly from its surround (in ways which can be characterized in terms of visual
attributes of the target and distractors), it will immediately draw attention towards it-
self. Thus, these experiments evidence how the composition of the visual scene alone
is a potentially very strong component of attentional control, guiding attention from
the bottom of the visual processing hierarchy up.