Perception and Attention - Computational Explorations in Cognitive Neuroscience - page 267

This model could be used to account for some of the
phenomena in the extensive literature on visual search
(e.g., Duncan & Humphreys, 1989; Treisman, 1996).
The typical visual search paradigm involves searching
for a target shape among a field of distractors. If the
target is distinct enough from the distractors, and if the
distractors themselves are relatively homogeneous, then
this search process can be quite rapid (i.e., it does not
appear to depend very much on the number of distrac-
tors). This rapid search has been described as taking
place in parallel. In contrast, when the distractors are
similar to the target, and/or there are multiple different
distractor shapes, search goes more slowly, and the time
seems to depend on the number of distractor stimuli.
This has been characterized as serial search (i.e., search-
ing sequentially through the display items until the tar-
get is located). In reality, the simple parallel/serial dis-
tinction appears to be more of a continuum that depends
on a number of factors (e.g., Duncan & Humphreys,
1989). See Mozer and Sitton (1998) for a visual search
model that has some key properties in common with the
one we have been exploring, which addresses some of
these subtleties.
Finally, we can replicate the object-based attentional
cuing that we demonstrated in the previous model. As
before, we present one object in the center of the dis-
play, and then present two objects on either side, and
see if the network processes the cued object. We also
cue with the object that the network would not other-
wise choose to process on its own as before, so that we
can be sure that the cuing is driving the effect. We will
do this by working with the underlying process control
panel that runs the testing that we've been doing, giving
you a better sense of the “guts” of the system.

You will see the network process object number 17
and identify it correctly.

Then, Step again.
You will see two objects (12 and 17) presented (this is
just the first event from the prior explorations). The net-
work processes the cued object this time, even though
in previous cases it has processed object 12 preferen-
tially. Note that, to ensure that the spatial representa-
tions do not contribute to the effect, we have reinitial-
ized them between the cue and subsequent inputs. Thus,
this demonstration shows that object-based attentional
cuing effects can be found even in this more realistic
object recognition model.

To stop now, quit by selecting Object/Quit in the
PDP++Root window.

8.6.2 Summary and Discussion

This model shows that the principles we developed re-
garding the interaction between spatial and object at-
tention scale up well to a more complex and realistic
model of object recognition. More generally, this model
demonstrates how spatial attention can help solve the
binding problem by restricting object-recognition pro-
cessing to only one object's features at a time. Spa-
tial attention is effective in this case because objects are
spatially contiguous.

8.7 Summary

The visual system entails a sequence of processing
stages, starting in the retina, continuing to the LGN
of the thalamus , and into area V1 in the cortex. V1
has topographically organized retinotopic representa-
tions that encode the visual world according to oriented
edges , providing a relatively efficient means of encod-
ing the visual form of objects. These V1 receptive field
properties arise naturally in a self-organizing model us-
ing Hebbian learning on natural visual scenes, and can
thus be understood as representing the basic correla-
tional structure present in visual scenes.
Beyond V1, visual processing is divided into two ma-
jor pathways. The object recognition or “what” path-
way involves areas in the ventral stream of processing

Do View , TEST_PROCESS_CTRL on the over-
all ( objrec_ctrl ) control panel, to bring up the
SequenceEpoch process control panel that controls the
testing process. Change the environment field on
the process control panel. Just click on the current
value (should be vis_sim_test ), and then go down
to the bottom of the menu under ScriptEnv , where
you should find obj_attn_test . Apply this change,
and then ReInit (equivalent to InitTest ) and Step
(equivalent to StepTest ).