Boston-NeuroTalks Calendar 3.0

Ken Nakayama (Harvard University)

Abstract:

The concept of the receptive field in visual science has been
transformative, fueling discoveries of the second half of the 20th century,
and supplying the dominant understanding of how the early visual
system works. Its reign has been extended to high level vision where in the
form of linear classifiers, it provides a framework to understand object
recognition. Untamed, however, are areas of visual perception, dubbed
variously as the 2.5 D sketch, mid-level vision, surface representations.

So far, neurons with their receptive fields seem unable to bridge the
explanatory gap, to supply us with even a speculative framework to
understand border ownership, amodal completion, subjective contours and a
variety of other surface phenomena. Lacking such theoretical support, these
areas have languished.

In reviewing this area, we single out occluding contours. These inferred
boundaries of objects and surfaces have a special status in providing
information about the geometrical relations of surfaces and objects in the
real world. Even though such important contours are not obviously evident
to observers in the clutter of surface markings of objects and terrain,
they have a special status in visual processing. A property of such
bounding contours which distinguishes them from all other contours is their
one-sided relation to surfaces. At the boundary defined by the contour,
only one side “owns” it. Border ownership is thus an inherent property of
bounding contours.

Ecological considerations provide a framework to explore the determinants
of bounding contours. This includes: (1) their specification by depth, (2) 
by convexity or smallness, and (3) entropy as a signal for background. In
addition, line drawings, because they tap into a primitive visual code,
provide additional evidence for the primacy of bounding contours.

In this context, subjective contours are particularly noteworthy because
they provide very precise evidence for the computation of occluding
contours. They are visible evidence for an otherwise important hidden
process. Rather than being relegated to mere illusion (i.e. error) or
having the minor role of contour completion, subjective contours reflect a
fundamental process of vision. As such, they provide a litmus test for
occluding contour computation.

Thus, phenomenology, rather than just describing the appearance of things,
reveals deeper substrates of vision. Accordingly, perceptual demonstrations
will be used support this interpretative framework.

http://bcs.mit.edu/newsevents/calendar.php?calendar=single&id=14974475