272 R. M. SHAPLEY AND C.
ENROTH-CUGELL
adaptation is the basis of brightness constancy. If
retinal adaptation averaged light " o n the entire
retina" as Mach supposed, it could not produce
brightness constancy when the illumination was
spatially non-uniform. Retinal gain would be the
same at all points on the retina. Neural signals at
one point on the retina would be attenuated to the
same extent as neural signals from another region
of unequal local illumination. The correct
calculation of the border contrast by the retina
would break down. This implies that the retinal gain
control mechanism must be somewhat localized, as
indeed physiological evidence d e m o n s t r a t e s
(Cleland and Enroth-Cugell, 1968 among others).
Mach and many others later, including Land and
McCann (1971), proposed that the visual system
calculated the ratio IRo/IRB = Ro/RB directly,
but this is incorrect, probably for the following
reason. The ratio R o / R s is nearly always close to 1
in nature because the (achromatic) reflectances of
most natural objects are so nearly the same. If
Ro/RB were the quantity being measured, correct
identification of whether an object is brighter or
darker than the background would become a
problem of accurate measurement and comparison
of the ratio with 1. The Weber contrast as defined
earlier is the difference between the reflectance ratio
and 1. That is, (Ro-RB)/RB = (Ro/Ra)-l. The
Weber contrast, (Ro-RB)/R B, changes sign for
objects brighter than the background (positive
contrast) compared to objects darker than the
background (negative contrast). The change of sign
leads to a m u c h easier, less e r r o r - p r o n e
discrimination of dark from bright objects than
would computation of the reflectance ratio Ro/RB.
Furthermore, the visual systems of vertebrates put
the basic measurement of the sign of the contrast
into the functional architecture of the retina (cf.
Hering, 1920). Cells which are excited by positive
contrast (the " o n " cells of Hartline, 1938; and the
" o n - c e n t e r " cells of Kuffler, 1953) are segregated
from the cells which are excited by negative contrast
( " o f f " cells of Hartline and " o f f - c e n t e r " cells of
Kuffler). As discussed in Appendix 1, the
elaboration of these " o n " and " o f f " pathways
involves separation of retinal synaptic connections
in the inner plexiform layer of the retina, as
suggested by Famiglietti and Kolb (1976), and then
proven by Famiglietti et al. (1977) and Nelson et
al. (1978). These physiological and anatomical
observations reinforce the purely functional
hypothesis that the retina is designed to measure the
contrast of objects in order to provide to the brain
an illumination-invariant description of the world
of objects. Spatially localized light adaptation is a
crucial factor in this retinal function.
1.2.2. TO HANDLETHE LARGERANGEOF
ILLUMINATIONLEVELS
A related reason for the necessity of light
adaptation is the very extensive range of average
light levels presented to the eye by nature. A white
paper (reflectance = 1) in moonlight has a
luminance of about 3 x 10-2 cd m-L A white paper
in sunlight is six orders of magnitude brighter,
about 3 x 104 cd m-L Backgrounds which affect
vision extend three log units below reflected
moonlight and one log unit above reflected sunlight,
a total range of about 10'% Part of this enormous
range is taken care of by parallel processing in
separate rod and cone pathways. In humans for
example, the highly sensitive rod system handles the
three lowest decades of backgrounds. The less
sensitive cone system handles the upper six log units,
and the decade of order 0.1 cd m -2 is shared (Riggs,
1965). However, in other animals the overlap of the
ranges of background handled by rod and cone
systems may differ because of a different rod - cone
weighting. For instance, in the cat in which the
rod - cone ratio is about one hundred times greater
than in man (Steinberg et al., 1973), the rod system
handles five log units rather than three and the
cones only dominate visual responses above
10 cd m -2 in background. A table which expresses
the luminances in standard units and in terms of
quanta of light per second per degree squared in
area, is offered as Table 1.
A human is so sensitive when dark adapted that
he can detect (without any false positive responses)
that a light flash has been presented when only
about one hundred quanta of light are incident on
his cornea (Nagel, 1909; Hecht et al., 1942; cf.
Cornsweet, 1970). When one takes into account
losses in the eye and the inefficiency of
transduction, this implies that about twenty retinal
responses to quanta of light are required for such
ultra-reliable visual performance (this is a higher
number than estimated by Hecht et al., 1942, and