VISUAL ADAPTATION AND RETINAL GAIN CONTROLS
meaning about mechanisms, it does reveal that
functionally the two processes seem to affect time
course of responses in similar ways.
Note that in none of these models is response
simply a gain factor times an unadapted signal.
Rather, in each case, adjustment of gain and
adjustment of time constants, a n d / o r strength of
negative feedback, appear to be necessary, in order
to explain the association of gain and time course
in the data from visual neurons. The response of
retinal neurons is thus a functional of the light now
and the past history of illumination. The retinal
functional is under the control of mean level and
mean contrast.
FIo. 61. A diagram of a model for the way the contrast
gaincontrol modifies first order responses of ganglion cells.
I c is the contrast signal. The model consists of a gain stage
A, unaffected by contrast, N L stages of low pass filtering,
and one stage of high pass filtering. Each low pass stage has
a time constant "r,. The high pass stage is a feedback loop
with one stage with a time constant TH, and with feedback
gain K. Contrast mainly affects K and TH. From Shapley
and Victor (1981).
6.2. Retinal Gain and Visual Sensitivity
The psychophysical consequences of retinal gain
control, and the implications of psychophysical
results about retinal gain controls, require for their
interpretation a theory linking the psychophysics
and neurophysiology of vision. In a previous
publication (Enroth-Cugell and Shapley, 1973b) we
suggested a sketch o f such a theory and here will
make it explicit.
The aims of the theory are: (i) to show how gain
and noise interact in determining sensitivity in rod
vision; (ii) to show how the dependence of gain
335
setting on receptive field size can be used to explain
the observed spatial effects on the psychophysical
laws of light adaptation (Barlow, 1957, 1965; van
Nes and Bouman, 1967; Koenderink et al. 1978).
In our proposed model, psychophysical threshold
is presumed to be governed by a signal-to-noise
ratio. The signal is presumed to be the neural
response of a population of retinal ganglion cells;
the noise is the variability of firing in those cells.
The signal is subject to gain control which is
supposed, in the theory, to behave in the way that
the retinal gain control has been observed to behave
in cat retinal ganglion cells. We presume that there
are a few (at least) retinal spatial channels at any
particular retinal locus. We postulate that rod and
cone signals are independent, in their noise
generation and in their gain control, until late in
the retinal stages of signal processing. Rod and cone
pathways must have independent dark noises, and
different dark adapted gains. There are two main
factors which determine the sensitivity of ganglion
cells: gain and noise (Barlow and Levick, 1969;
Rose, 1948). From work cited above, the gain of
a ganglion cell's response to a step of light on a
background in the scotopic range, G R is:
G R = GRO/(1 + FB/FRo)P (37)
where GR0 is the dark adapted gain, FB is the
background flux effectively absorbed, FRO is the
criterion amount of flux which must be exceeded
to turn on the gain control of the rod pathway, and
P is an exponent, usually measured to be 0.9, which
we will approximate to 1 (Cleland and Enroth-
Cugell, 1970; Enroth-Cugell and Shapley, 1973a).
The noise will be the dark noise summed together
with the background flux over the receptive field.
Thus the variance o 2 of the noise from the rod path-
way in a ganglion cell's activity will be
oR2 -- [ad "(FB + FRo)] (38)
FRD = IRD • fll t
FB = IB " A ,
Where IB is the background retinal illumination,
A, is the summing area of the ganglion cell center,
FRD is the dark noise in the rod pathway (in
equivalent quanta s-X). It is assumed that the