Eye, Brain, and Vision

the sweeping had to be done with the edge in one particular orientation. Most
amazing was the contrast between the machine-gun discharge when the orien-
tation of the stimulus was just right and the utter lack of a response if we
changed the orientation or simply shined a bright flashlight into the cat's eyes.
The discovery was just the beginning, and for some time we were very
confused because, as luck would have it, the cell was of a type that we came
later to call complex, and it lay two stages beyond the initial, center-surround
cortical stage. Although complex cells are the commonest type in the striate
cortex, they are hard to comprehend if you haven't seen the intervening type.
Beyond the first, center-surround stage, cells in the monkey cortex indeed
respond in a radically different way. Small spots generally produce weak re-
sponses or none. To evoke a response, we first have to find the appropriate
part of the visual field to stimulate, that is, the appropriate part of the screen
that the animal is facing: we have to find the receptive field of the cell. It then
turns out that the most effective way to influence a cell is to sweep some kind
of line across the receptive field, in a direction perpendicular to the line's orien-
tation. The line can be light on a dark background (a slit) or a dark bar on a
white background or an edge boundary between dark and light. Some cells
prefer one of these stimuli over the other two, often very strongly; others
respond about equally well to all three types of stimuli. What is critical is the
orientation of the line: a typical cell responds best to some optimum stimulus
orientation; the response, measured in the number of impulses as the receptive
field is crossed, falls off over about 10 to 20 degrees to either side of the
optimum, and outside that range it declines steeply to zero (see the illustration
on the facing page). A range of 10 to 20 degrees may seem imprecise, until you
remember that the difference between one o'clock and two o'clock is 30 de-
grees. A typical orientation-selective cell does not respond at all when the line
is oriented 90 degrees to the optimal.
Unlike cells at earlier stages in the visual path, these orientation-specific cells
respond far better to a moving than to a stationary line. That is why, in the
diagram on the facing page, we stimulate by sweeping the line over the recep-
tive field. Flashing a stationary line on and off often evokes weak responses,
and when it does, we find that the preferred orientation is always the same as
when the line is moved.
In many cells, perhaps one-fifth of the population, moving the stimulus
brings out another kind of specific response. Instead of firing equally well to
both movements, back and forth, many cells will consistently respond better
to one of the two directions. One movement may even produce a strong
response and the reverse movement none or almost none, as illustrated m the
figure on the facing page.
In a single experiment we can test the responses of 200 to 300 cells simply by
learning all about one cell and then pushing the electrode ahead to the next cell
to study it. Because once you have inserted the delicate electrode you obvi-
ously can't move it sideways without destroying it or the even more delicate
cortex, this technique limits your examination to cells lying in a straight line.
Fifty cells per millimeter of penetration is about the maximum we can get with
present methods. When the orientation preferences of a few hundred or a
thousand cells are examined, all orientations turn out to be about equally rep-
resented—vertical, horizontal, and every possible oblique. Considering the
nature of the world we look at, containing as it does trees and horizons, the
question arises whether any particular orientations, such as vertical and hori-
zontal, are better represented than the others. Answers differ with different
laboratory results, but everyone agrees that if biases do exist, they must be
small—small enough to require statistics to discern them, which may mean
they are negligible!
In the monkey striate cortex, about 70 to 80 percent of cells have this prop-
erty of orientation specificity. In the cat, all cortical cells seem to be orientation
selective, even those with direct geniculate input.
We find striking differences among orientation-specific cells, not just in op-
timum stimulus orientation or in the position of the receptive field on the
retina, but in the way cells behave. The most useful distinction is between two
classes of cells: simple and complex. As their names suggest, the two types differ
in the complexity of their behavior, and we make the reasonable assumption
that the cells with the simpler behavior are closer in the circuit to the input of
the cortex.

SIMPLE CELLS
For the most part, we can predict the responses of simple cells to
complicated shapes from their responses to small-spot stimuli. Like retinal
ganglion cells, geniculate cells, and circularly symmetric cortical cells, each
simple cell has a small, clearly delineated receptive field within which a small
spot of light produces either on or off responses, depending on where in the
field the spot falls. The difference between these cells and cells at earlier levels
is in the geometry of the maps of excitation and inhibition. Cells at earlier
stages have maps with circular symmetry, consisting of one region, on or off,
surrounded by the opponent region, off or on. Cortical simple cells are more
complicated. The excitatory and inhibitory domains are always separated by a
straight line or by two parallel lines, as shown in the three drawings on this
page. Of the various possibilities, the most common is the one in which a
long, narrow region giving excitation is flanked on both sides by larger re-
gions giving inhibition, as shown in the first drawing (a).
To test the predictive value of the maps made with small spots, we can now
try other shapes. We soon learn that the more of a region a stimulus fills, the
stronger is the resultant excitation or inhibition; that is, we find spatial summa-
tion of effects. We also find antagonism, in which we get a mutual cancellation
of responses on stimulating two opposing regions at the same time. Thus for a
cell with a receptive-field map like that shown in the first drawing (a), a long,
narrow slit is the most potent stimulus, provided it is positioned and oriented
so as to cover the excitatory part of the field without invading the inhibitory
part (see the illustration on the facing page). Even the slightest misorientation
causes the slit to miss some of the excitatory area and to invade the antagonistic
inhibitory part, with a consequent decline in response.
In the second and third figures (b and c) of the diagram on this page, we see
two other kinds of simple cells: these respond best to dark lines and to dark/
light edges, with the same sensitivity to the orientation of the stimulus. For all
three types, diffuse light evokes no response at all. The mutual cancellation is
obviously very precise, reminiscent of the acid-base titrations we all did in
high school chemistry labs. Already, then, we can see a marked diversity in
cortical cells. Among simple cells, we find three or four different geometries,
for each of which we find every possible orientation and all possible visual-
field positions.
The size of a simple-cell receptive field depends on its position in the retina
relative to the fovea, but even in a given part of the retina, we find some
variation in size. The smallest fields, in and near the fovea, are about one-
quarter degree by one-quarter degree in total size; for a cell of the type shown
in diagrams a or b in the figure on this page, the center region has a width of
as little as a few minutes of arc. This is the same as the diameters of the
smallest receptive-field centers in retinal ganglion cells or geniculate cells. In
the far retinal periphery, simple-cell receptive fields can be about 1 degree by 1
degree.
Even after twenty years we still do not know how the inputs to cortical cells
are wired in order to bring about this behavior. Several plausible circuits have
been proposed, and it may well be that one of them, or several in combination,
will turn out to be correct. Simple cells must be built up from the antecedent
cells with circular fields; by far the simplest proposal is that a simple cell
receives direct excitatory input from many cells at the previous stage, cells
whose receptive-field centers are distributed along a line in the visual field, as
shown in the diagram on the next page.
It seems slightly more difficult to wire up a cell that is selectively responsive
to edges, as shown in the third drawing (c) on the facing page. One workable
scheme would be to have the cell receive inputs from two sets of antecedent
cells having their field centers arranged on opposite sides of a line, on-center
cells on one side, off-center cells on the other, all making excitatory connec-
tions. In all these proposed circuits, excitatory input from an off-center cell is
logically equivalent to inhibitory input from an on-center cell, provided we
assume that the off-center cell is spontaneously active.
Working out the exact mechanism for building up simple cells will not be
easy. For any one cell we need to know what kinds of cells feed in informa-
tion—for example, the details of their receptive fields, including position, ori-
entation if any, and whether on or off center—and whether they supply excita-