Image Processing Reference
In-Depth Information
Time
Off
Off
On
Time
Time
Fig. 1.7. The graph on the
left
illustrates that the complex cells of V1 are insensitive to bar
position. On the
right, top
and
right, bottom
the responses of motion-direction sensitive and a
motion-direction insensitive complex cell responses are shown
tracking of moving objects. Except for those which are of high-pass type, the opti-
mal velocity of velocity-tuned cells increases with visual eccentricity and appears to
range from 2
◦
to 90
◦
per second. To limit the scope of this topic and also because
they are less studied, we will not discuss cells beyond area V1 further, and refer to
further readings, e.g., [173].
Complex cells are encountered later in the computational processing chain of
visual signals than are simple cells. Accordingly, to construct their outputs, the com-
plex cells presumably receive the outputs of many simple cells as inputs. As in the
case of simple cells, the exact architecture of input-output wiring of complex cells
has not been established experimentally, but there exist suggested schemes that are
being debated.
There is repeatedly convincing evidence, e.g., [4, 6, 45, 46, 159, 165], suggesting
the existence of well-organized cells in V1 that exhibit a spatial frequency selec-
tivity to moving and/or still sinusoidal gratings, e.g., the top left of Fig. 10.2. The
cells serving fovea in V1, have optima in the range of 0.25-4 cycles/degree and have
bandwiths of approximately 1.5 octaves [165]. Although these limits vary somewhat
between the different studies that have reported on frequency selectivity, even their
very existence is important. It supports the view that the brain analyzes the visual
stimuli by exploding the original data via frequency, spatial direction, and spatio
temporal direction (velocity) channels in parallel before it actually reduces and sim-
plifies them, e.g., to yield a recognition of an object or to generate motor responses
such as those of catching a fast ball.
