Transcript Phenomenal Coherence of Moving Visual Patterns

Phenomenal Coherence of
Moving Visual Patterns
Edward Adelson & Anthony Movshon
Question
Is the aperture problem solved by a two-step
system of motion detection?
Alternatives

Yes, the two-step model is the solution.

No: Its somehow solved otherwise

No: Its not a problem to begin with
Logic

If there are two levels of motion detection, we might
be able to create some stimuli in which the second
level is active and some in which it is not.

The activity of the second level is apparent in stimuli
where the aperture problem is resolved.

By studying the properties of the stimuli that are
perceived as resolved vs. nonresolved, we can infer
the properties of the second level.
Methods

Subjects were presented with “plaid” stimuli
that consisted of two moving sinusoidal
gratings.

One of the gratings varied systematically in
contrast and spatial frequency.

Subjects were asked to indicate if the percept
was coherent or transparent.
Methods – details



Subjects were asked to not make a judgment
unless their eyes were fixated on a small
black circle in the middle of the display.
After each 1.5-s exposure, the observer
indicated whether the pattern appeared
‘coherent’ or ‘incoherent’.
Stimuli were 5 deg, presented on a CRT,
controlled by a PDP11
Results


Two gratings that are sliding
over each other are often
seen as moving coherently
in one, not two directions
(b).
This common direction is
best described as the
“intersection of constraints”
not the vector average (c).
Results



Increasing the contrast of one
of the gratings makes it more
likely that it is perceived (open
circles)
Increasing the contrast further
makes the percept
monotonically more and more
coherent.
 Contrast affects coherence
beyond mere visibility.
Results



Utilizing this contrast
dependence, the authors
recorded under which conditions
of “added” contrast, they could
get 50% coherence.
As it turns out, the larger the
difference in spatial frequency
between the gratings is, the
more contrast needs to be
added.
 Mechanism of coherence
detection is SF dependent.
Interpretation



This leads to a two-step model:
At a first level, oriented and SF-dependent neurons in V1
sample the local motion of an object, extracting the motion
perpendicular to the contours.
At a second level, this information is integrated by MT
neurons that perform a Intersection of Contraints
computation on the component vectors.
Problems
1.
2.
3.
The data is consistent with this notion, but
is so indirect that it does not offer concrete
proof that MT plays a part in second stage
processing.
Experiments done in human whereas all
physiology on MT done in monkeys (as of
1982).
There were only two subjects