| Demo 13.10. Contour segmentation process.
This animated version of Figure
13.10 shows how the neurons in the PGLISSOM orientation map
synchronize and desynchronize their spiking activity to represent
different contours. The input presented to the network is shown in
gray-scale coding at left, the areas of the map that respond to the
different input elements are delineated with circles in the middle,
and the neural spiking in the 54 × 54 GMAP is shown as black and
white dots at right (black means the neuron is spiking at the current
time step, white means that it is not spiking). Each contour was
composed of three contour elements (numbered 1, 2, 3 and 4, 5, 6),
embedded in a background of three randomly oriented elements.
PGLISSOM performs contour integration through synchronized and
desynchronized neural activation: Neurons that represent elements of
the same contour spike at the same time, and those that represent
elements in different contours spike at different times. Through
self-organization, principles of good continuation and proximity have
become encoded in the excitatory lateral connections, i.e. neurons
that represent collinear or co-circular paths tend to be connected.
The lateral connections mediate synchronization, and as a result,
PGLISSOM groups collinear and co-circular elements together into
continuous contours.
The neurons representing each contour quickly synchronize their
activity. On the other hand, the neurons representing elements in
different contours, of elements in the background, and of contour and
background elements become desynchronized. In other words, the three
areas representing the same contour fire together while the areas
responding to the other contour and to the background are silent.
(The grouping is most visible when the animation is speeded up to 0
delay). Such an alternating activation of neuronal groups ensures that
each coherent object is represented distinctly and not mixed with
representations of other objects.
Previous demo;
Next demo
|
