Modelling Primate Visual Attention - Computational Neuroscience: A Comprehensive Approach - page 611

the next location stored in the sequence being tested against the new scene), hidden
Markov models (where sequences are stored as transition probabilities between lo-
cations augmented by the visual features expected at those locations), or evidential
reasoning (similar to the model of Schill and colleagues). These models typically
demonstrate strong ability to recognize complex grayscale scenes and faces, in a
translation, rotation and scale independent manner, but cannot account for non-linear
image transformations (e.g., three-dimensional viewpoint change).
While these models provide very interesting examples of cooperation between a
fast attentional cueing system and a slower localized feature analysis system, their
relationship to biology has not been emphasized beyond the general architectural
level. Teasing apart the brain mechanisms by which attention, localized recognition,
and rapid computation of scene gist and layout collaborate in normal vision remains
one of the most exciting challenges for modern visual neuroscience [40].

19.6.3 Attention as a component of vision

In this section, we have seen how vision relies not only on the attentional subsystem,
but more broadly on a cooperation between crude preattentive subsystems for the
computation of gist, layout and for bottom-up attentional control, and the attentive
subsystem coupled with the localized object recognition subsystem to obtain fine
details at various locations in the scene ( Figure 19.4 ) .
This view on the visual system raises a number of questions which remain fairly
controversial. These are issues of the internal representation of scenes and objects
(e.g., view-based versus three-dimensional models, or a cooperation between both),
and of the level of detail with which scenes are stored in memory for later recall and
comparison to new scenes (e.g., snapshots versus crude structural models). Many
of these issues extend well beyond the scope of the present discussion of selective
visual attention. Nevertheless, it is important to think of attention within the broader
framework of vision and scene understanding, as this allows us to delegate some of
the visual functions to non-attentional subsystems.

19.7 Discussion

We have reviewed some of the key aspects of selective visual attention, and how
these contribute more broadly to our visual experience and unique ability to rapidly
comprehend complex visual scenes.
Looking at the evidence accumulated so far on the brain areas involved with the
control of attention has revealed a complex interconnected network, which spans
from the earliest stages of visual processing up to prefrontal cortical areas. To a large
extent, this network serves not only the function of guiding attention, but is shared
with other subsystems, including the guidance of eye movements, the computation