Transparent Motion Perception as Detection of Unbalanced Motion
Signals III: Modeling
Ning Qian, Richard A. Andersen and Edward H. Adelson
Published in
Journal of Neuroscience, vol. 14, pp. 7381-7392 (1994).
In the preceding two companion articles we studied the conditions
under which transparent motion perception occurs through psychophysical
experiments, and investigated the underlining neural mechanisms through
physiological recordings. The main finding of our perceptual experiments
was that whenever a display has finely balanced motion signals in all
local areas, it is perceptually nontransparent, and that transparent
displays always contain motion signals in different directions that are
either spatially unbalanced, or unbalanced in their disparity or spatial
frequency contents. In the physiological experiments, we found two
stages in the processing of transparent stimuli. The first stage is
located primarily in area V1. At this stage motion measurements are made
and V1 cells respond well to both the balanced, nontransparent stimuli
and the unbalanced, perceptually transparent stimuli. The second stage
is located primarily in area MT. MT cells show strong suppression
between opposite directions of motion. The suppression for the
unbalanced, transparent stimuli is significantly less than that for the
balanced, nontransparent stimuli. Therefore, the activity in the second,
MT stage correlates better with the perception of motion transparency
than the first, V1 stage, which does not distinguish reliably between
transparent and nontransparent motion.
The above experiments suggest a two-stage model of motion perception
with a motion measurement stage in V1 and an opponent-direction
suppression stage in area MT. In this article we explicitly test this
model through analysis and computer simulations, and compare the
response of the model to the perceptual and physiological results using
the same balanced and unbalanced stimuli we used in the experiments. In
the first stage of the computational model, motion energies in different
spatial frequency and disparity ranges are extracted from each local
region. Similar to V1, this stage does not distinguish between the
balanced and unbalanced stimuli. In the subsequent stage motion energies
of opposite directions but with same spatial frequency and
disparity contents suppress each other using subtractive or divisive
inhibition. This stage responds significantly better to the transparent
stimuli than to the nontransparent ones, in agreement with MT activity.