Problems posed by changing illumination levels, and the impact of light level on vision.

1
Night and Day
Problems posed by changing illumination levels,
and the impact of light level on vision.
2
1. Night Vision
The Photon nature of light poses a special
problem for seeing at night.
3
Photons arrive individually!
Interesting calculation from The First Steps
in seeing by RW Rodieck.
When looking at Polaris (North Star) on a clear
night, 2,350 visible photons illuminate each mm2
of the eyes pupil each second. At night the
pupil area is about 40 mm2, thus about 94,000
visible photons pass through the pupil, and of
these 55,000 arrive at the photoreceptors per
second. Since the eyes shutter speed is about
0.1 seconds, 5,500 photons make up the retinal
image of Polaris during 0.1 seconds. Since the
velocity of light is 300,000 km/sec, the 5,500
photons that make up the image during 0.1 seconds
form a column of photons 30,000 km long, thus
each photon is on average about 5 km (3 miles)
apart.
4
How are these photons distributed across the
retina?
Photon distribution in the foveal image of Polaris
Central 50 photons
Central 25 photons
5
Fate of Photons arriving at photoreceptors 1.
With 7 mm pupil, only 54 of 5,500 photons (2,900
photons) line up with receptor axis and are
funneled by the inner segments into the
photopigment rich outer segments and thus have
the potential to be absorbed by photopigment. 2.
Of the 2,900 photon passing through the
outer-segments of the cones, 700 are absorbed and
produce photo-isomerizations (25 quantum
efficiency). Of these, only about 50 (350
photons) form the central peak in the image of
Polaris.
6
Very Dim Lights are seen by individual rods
responding to single photo-isomerizations!
.
.
.
.
.
.
.
.
.
.
Viewing a dark sky over the tops of some trees at
night.
Trees 83 photons
Sky 280 photons
The chance that any single rod would absorb more
than 1 photon is very small, thus, since we can
see this boundary, must be seeing with just
single photoisomerizations in any rod.
7
How many photo-isomerizations are necessary to
see?
An example The rod system integrates (sums)
photons over about 0.1 second. We can think of
this as the shutter speed of human scotopic
vision. Dark night sky In the retinal image
there are 500 photons/mm2/0.1 second. Since we
have about 160,000 rods per mm2, this means that
1 photon is absorbed by each rod every 30
seconds. Or, for any 0.1 second period, only 1
in 300 rods captures a photon.
8
What limits the minimum retinal illuminance?
Dark Light (Eigengrau) or thermal
isomerizations. Dark Light Equivalent to 1000
photons per mm2/second, Or 100 photons/ mm2/0.1
second. Thus this is equivalent to one photon
being absorbed by each rod every 160
seconds. Since there are 1.4 x 108 rhodopsin
molecules per rod outer segment, this means that,
on average, each rhodopsin molecule isomerizes
spontaneously (without absorbing a photon)
every 710 years. Very stable photopigment!
9
2 Gain Control
Decreasing Gain to prevent saturation sometimes
called Light Adaptation
10
Stars are easy to see at night and invisible
during the day, while trees are hard to see at
night and highly visible during the day. Stars
are emitting sources, while trees are reflective
sources.
.
Luminance
DAY
Night
.
.
.
.
.
.
.
Night
.
.
DLk
star
sky
DAY
DLkL or DL/Lk
sky
Tree
11
Two sources of noise limit sensitivity to very
dim lights 1. Dark Light (thermal
isomerizations) is random and constant. 2.
Photon noise Light is Poisson Distributed, thus
VAR mean. If VAR mean, then SD square root
of mean.
Increasing light levels make emitting light
sources such as stars more difficult to see, but
make reflective targets easier to see simply
because of photon noise. Emitting source at
high light levels signal is same (DL), but noise
is higher, so more difficult to see. Reflective
source Signal intensity (DL) increases in direct
proportion to light level, but photon noise (SD)
only increases with square root of light level,
so signal/noise ratio increases as light levels
go up and thus reflective targets become more
visible at high light levels.
12
The impact of photon noise on visibility Remember
that detectability (d) is Dmean/SD.
1. Emitting source such as a star DLk
70,002
High Light level 10,000 higher illum
9
Low light
70,000
7
d 2/sqrt(7) 0.75
d 2/sqrt (70,000) 2/264 0.0075
2. Reflecting source such as a piece of paper
DLkL
90,000
9
70,000
7
d 2/sqrt(7) 0.75
d 20,000/sqrt (70,000) 20,000/264 67
If photon noise was the only determinant of
visibility, as light levels increase by 10,000
times, stars become 100 times less visible, but
reflective targets become 100 times more visible.
This is called the DeVries-Rose or Square-root
Law (100sqrt10,000)
13
Image Intensification At low light Levels
1. Increase Photons in image Dilate pupil,
factor of 10 2. Employ fast film Switch to
rod vision. Rods are high gain Change in dark
current for single photon is 20X that of
individual cone 0.7 pA for rod, 0.03pA for
cone. Single rhodopsin molecule isomerized,
moves around membrane and changes about 700
G-proteins and reduces cG by 1400 (0.7), this is
disk amplification. Channel amplification 1400
cG molecules close 230 (2) ion channels, this
leads to a 0.7 pA reduction in dark current. 3.
Sum photons over larger area and longer duration
(camera analogy, shutter speed and size of silver
crystals)
14
Two amazing features of human vision related to
light level.
1. We can see objects when the ambient light
level is as low as 1/1,000,000 cd/m2 up to
1,000,000 cd/m2. That is, our visual system is
able to function over a 1012 range of light
level. This is very impressive! 2. Our
perception of objects is never confused by
changing light levels. That is, even though a
black piece of paper seen on a sunny day may
reflect 1,000,000 times more light than a white
piece of paper seen while reading under dim
light, the black paper never appears brighter
than the white paper and the white paper never
appears black. Brightness constancy is truly
amazing!
15
The explanations of point 1 and 2 are
interrelated. Although these properties of human
vision have been well documented for more than
100 years, we are still struggling to find a
complete explanation.
First, remember that Luminance is determined by
both object reflectance AND the illuminant
(L(ER)/pi). Also, remember that ideally,
vision would inform us about objects, and not be
corrupted by illuminants (remember the case of
color constancy). In this case, we want the
visual system response of brightness to
indicate object reflectance not object luminance
because object reflectance is a property of the
object and will not vary with illuminance level.
16
The need for Brightness Constancy.
Dusk
Noon
E10,000
E10
R0.5
R0.5
R0.8
R0.8
Lum (gray) 5000/pi Lum (white) 8000/pi
Lum (gray) 5/pi Lum (white) 8/pi
Notice that during the day the gray paper has a
luminance 5000/8 625 times higher than the
white paper at dusk. If brightness was
determined by the quantity of light (luminance),
then the gray paper at noon would appear to be a
brilliant white, which as we all know, it does
not. That is, if perception was based upon light
level, we would confuse white and gray paper.
17
Analogy of cameras and film speed.
Bright sunny day
Dimly lit room
Optical density of negative
ASA 800 fast film
ASA 64 slow film
Object Luminances
Even black paper on a sunny day will reflect more
light than white paper in a dimly lit room since
reflectance of black paper may be 5, and 5 of
10,000 is more than 85 (white) of 10. The low
light level film has to spread its response over
a low light level range, whereas slow daylight
film has to spread its response over a much
higher and wider range.
18
Plotting in log units
If white paper reflect 80 of the incident light
and a gray paper only 50, the difference in
their luminances will always be 30 of the
illumination level. That is the gray will always
reflect 5/8 of that reflected from the white.
This is a constant ratio and NOT a constant
difference in luminance. A constant ratio is a
constant number of log units, thus we often plot
these data on log axes.
Bright sunny day
Dimly lit room
Optical density of negative
ASA 64 slow film
ASA 800 fast film
Log luminance
19
The fact that we can see just fine under all of
these conditions indicates that we have
adjustable speed film in our retinas.
Bright sunny day
Dimly lit room
ASA 800 fast film
ASA 64 slow film
Object Luminance
Note that we refer to the slope of these
response curves as GAIN. High speed film is
really high GAIN film. It is called high speed
or fast film because the camera can reach the
desired film exposure by opening the shutter for
a shorter period of time. This is refered to as
shutter speed, and thus the name high speed
film.
20
The human visual system seems to work at all
light levels indicating that we can adjust
(adapt) the gain of our visual system to any
light level. This is referred to as Gain
Control or Light Adaptation. How do we do this?
Dimly lit room
Bright sunny day
Brightness
High gain
Low gain
Log Retinal Illuminance
21
Impact of illumination level on image when gain
is fixed.
Reflectance of object
Response saturation
Illuminance Level
Without gain control, all of the object looks
dark at low light levels, and all appears bright
at high levels. In a camera we describe this as
under-exposure and over-exposure.
Fixed gain response
response
Object luminance range
luminance
22
Gain Control solves two problems introduced at
the beginning 1. The response range is always
matched to the input range, therefore, even
though the input range is huge (1012), we do not
saturate or bottom out. 2. Since the response
range always matches the input luminance range,
responses to white and gray and black are the
same at all light levels. Thus responses
represent object reflectance and not object
luminance. This is brightness constancy.
dusk
noon
Low gain
response
High gain
input
23
At low light levels, the visual system needs
image intensification or amplification
At high light levels, the visual system needs
image attenuation
Beyond the retina, the visual system has a fixed
gain, thus gain control (light adaptation) must
happen in the retina or optics.
Fixed gain response
Post-retinal Response
Retinal input
24
Gain control in the human visual system Gain
control at high light levels needs to act like
dark glasses, or neutral density filter and act
in a way analogous to reducing light levels down
to those that are appropriate for the system
gain. Alternately, at low light levels, gain has
to be increased similar to light glasses that
will in effect increase the light level (image
intensification). BOTH of these factors are
functioning in human vision. 1. Pupil size (vary
aperture in camera) 2. Two speeds of
photo-receptors (rods and cones). 3. Vary
spatial summation (e.g. silver crystals in
photographic film) and temporal summation (vary
shutter speed in camera) 4. Bleach
photopigment (photon capture) 5. Adjust gain of
individual photoreceptors (not pigment
bleaching) 6. Adjust gain of individual
post-receptoral neurons (lateral inhibition)
25
How do Photoreceptors adjust gain? Rods and
cones are quite different. Cones can adjust
gain, but in rods gain is always set at a maximum
and because of this, rods, unlike cones,
saturate. In both cases there are
post-receptoral gain mechanisms active.
More than one isomerization summed
Absolute Scotopic threshold
Rod Response (change in Dark current)
Rod saturation
10-4
10-3
10-2
10-1
1
10
102
103
104
105
106
Photons captured per second
Over most of the scotopic range, individual rods
only absorb single photons within the 0.1 second
shutter time. Once more than one isomerization
occurs simultaneously, further increases in
isomerizations rapidly lead to saturation, that
is, the individual rods only have a small
dynamic range of about 50 to 100.
26
Could it be due to photopigment depletion or
bleaching? No, there are 140 million rhodopsin
molecules in the outer segment, and since
response saturation happens when less than 100
are isomerized, clearly, most remain able to
absorb a photon and isomers Has some other
limited resource been depleted? Due to
amplification, many more ion channels are closed
than rhodopsin molecules isomerized, and there
are far fewer of these. Very simply, it only
takes about 50 to 100 photoisomerizations to
close ALL of the open ion channels, and thus stop
all of the dark current. Since rods respond by
changing the dark current, there can be no
possible additional response to more
isometrizations.
What causes this response saturation?
Rod response
100
1
Isomerizations within Temporal summation window
(0.1 second)
27
Rod saturation
Rod dark current has dropped to zero
Different numbers of isomerizations
3 isomerization
Notice that the 30 pA dark current is reduced by
photoisomerizations, and with sufficient numbers
of these (approximately 100), the dark current
drops to zero. Notice that the neural response
has a latency of about 40 msec, and persists for
more than 0.5 seconds for the brighter flashes.
28
Is there any gain control in the rod system? Yes,
but not in the rods themselves.
In order to increase sensitivity, the scotopic
visual system integrates over quite large areas.
This cannot be achieved by making road very large
(for some reason), thus, spatial summation is
achieved by converging the responses of many rods
onto single bi-polar cells.
As with most neurons, the bipolars have a dynamic
range of perhaps 100. Thus, since their input
can range for a single isomerization in a single
rod, up to 100 isomerizations in 15 (or more)
rods, the bipolars need some form of gain control
to be able to respond (change their response)
over this range.
Gain control is achieved via lateral inhibition
from Horizontal cells. Horizontal cells have
large RF and they are all linked, thus, when the
ambient light level goes up, the H-cells all
receive increased input, and this leads to
increased inhibition of the bipolars, reducing
their responses. This is active neural gain
control, and it takes about 0.1 seconds to happen.
29
What are the consequences of this gain control in
outer-plexiform layer?
The scotopic post-receptor neural gain control
necessary to prevent the bipolars saturating
should manifest itself as a reduction in
sensitivity, and it does. Psychophysical
sensitivity to light increments begins to decline
at much lower light levels than those necessary
to saturate rods. Rod saturation and reduced
sensitivity begins at light levels that have
already decreased the visual sensitivity by 1,000
times.
30
Sensitivity and Gain Control in Cones
Once the rods have saturated, no amount of
post-receptor gain control can produce functional
vision. Thus, since rods are saturating at light
levels experienced at dusk or in a dimly lit
room, there is need for an additional inherent
low gain system. This supplemental system
(cones) does not need to be as sensitive as rods,
but it must not saturate at high light levels.
Cones can respond to single photons, but the cone
or photopic visual system cannot respond to such
low light levels as those detected by the
rod-based scotopic system because of reduced
spatial and temporal summation (shutter speed of
1/20 second) and increased dark light in cones.
However, the lowest intensity light that can be
detected by cones (3 photons/cone/second) only
produces single isomerizations in individual
cones within the 1/20 second shutter speed.
Perhaps because of cone noise, three cones must
simultaneously capture a photon for the light to
be seen. Result photopic absolute threshold
150,000 scotopic absolute threshold.
Maximum intensity that can be seen might be white
snow at noon at high altitude, this produces
about 1,000,000 photons per cone per second.
Thus the dynamic range of cones ranges from 1 to
50,000 photons per 1/20 second. How does the
cone achieve such a large dynamic range
(approximately 1000 times that of rods)? It must
have its own internal gain-control system. Cones
never saturate to steady lights before damage
occurs.
31
Cones do saturate at about 100 isomerizations to
a flashed stimulus (just same as rods), but never
to a steady light source. Gain control is not
instant. Mechanisms of gain control in
cones 1. Active gain control Enzyme
Guanylate cyclase will generate extra cG in
response to its decrease caused by
photo-isomerizations. Thus, as ion channels
start to close, they soon start to open again
thus preventing saturation of dark current. 2.
Bleaching Since 11-sis retinal regeneration is
slow (mean 2 minutes), at high light levels,
reduce photon capture by decreased availability
of 11-cis retinal acts as passive gain control.
E.g. at 1,000 photons absorbed per cone per 1/20
sec, produces 2,000,000 bleached rhodopsin
molecules in 2 minutes
Active gain control opening up closed ion channels
32
Response of cones to a steady light does not
saturate because gain-control quickly reduces the
response after about 100 msec.
Gain control
Cone response
Negative after-image produced by gain control
light
time
33
Psychophysical sensitivity results of gain
control and image intensification
TVI curve
5
Log DI
Webers Law DIkI1 logDIk(Ix1)
4
3
deVries-Rose Law DIksqrt(I) kI0.5 Log DI
k(Ix0.5)
DI
2
I
1
Log Background Intensity (I)
1. Slope zero, dark light 2. Slope 0.5,
photon noise (deVries-Rose law) 3. Slope 1,
Webers law, post receptoral gain control 4.
Slope infinite, rod saturations (not typically
seen) 5. Slope 1, cone Weber region
34
Pathology that affects dynamic range of vision.
1. Congenital stationary Night Blindness Lack
protein that de-activates rod rhodopsin, this
activated rhodopsin remains around and continues
to closed ion channels, thus rods saturate at
very low light levels, and patients cannot see at
typical scotopic light levels. 2. Retinitis
Pigmentosa (heterogeneous group of
diseases) Slowly rod rhodopsin becomes
dysfunctional, and patients have complete
night-blindness.
35
Is gain control global or local? 1. The human
visual system local gain control , thus get
after images and each region of the retina
self-adjusts providing appropriate gain for each
region of the visual field. 2. Video Camera
Gloabal gain control. No after images and
suffers from saturation in unevenly illuminated
environments.
36
After Images
After-images evidence for local grain control.
37
(No Transcript)
38
Seeing bleaching, photochemistry, and gain
control at work through afterimages
Demo

• Intense Flash Get full neural saturation (do
  not see any detail, just bright white), but not
  full PR bleach.
• Positive After image, lingering photoreceptor
  response to activated and bleached photopigment,
  which indicates that the biochemical deactivation
  mechanisms are imperfect
• Negative after-image with same opposites as seen
  by Hering, indication of the reduced gain due to
  light adapted retina following the bleach

Question What is your VA with an after-image?
Can you see 20/20 without any light? How local
is gain control?
39
Dark Adaptation
The slow process of increasing the gain after
prolonged exposure to high light levels.
40
Photopic and Scotopic Spectral Sensitivity Curves
5
4
Notice that Scotopic and Photopic vision is
similarly sensitive at wavelengths gt650 nm
Log Relative Threshold
555nm
3
Photo- chromatic Interval
2
505nm
1
400
500
600
700
Wavelength (nm)
41
Higher adaptation levels produce 1. Greater
desensitization 2. Longer to reach full dark
adaptation 3. Visible rod-Cone break
8
7
6
5
6 log unit change in threshold (1 million
times!!)
Log Relative Threshold
Rod-cone break
4
3
2
1
0
10
20
30
40
Time in the dark (minutes)
42
Do not see rod-cone break when 1. There is no
significant cone bleaching 2. There are no rods
(e.g. test with small foveal target) 3. Test
with long stimulus wavelength (650 nm ) where
rods are not really any more sensitive that cones
8
7
6
Foveal or red test
5
Log Relative Threshold
4
Blue/green test in perifovea after very intense
bleach
3
Only rod bleaching
2
1
0
10
20
30
40
Time in the dark (minutes)
43
What is happening during Dark Adaptation in rods?
Retinal Densitometry (ratio of reflected light/
reflected light when fully bleached) shows the
proportion of bleached pigment in the rods. As
more pigment is regenerated, more light is
absorbed, and thus less reflected. Analogous to
measuring the brightness of the fundus over time
at the wavelength maximally absorbed by rhodopsin.
Proportion of bleached rhodopsin
Time (minutes) after full bleach
Notice that this exponential decrease in
bleached rhodopsin parallels the decrease in log
threshold seen in DA curves. Thus William
Rushton concluded that log visual threshold is
proportional to the proportion of bleached
rhodopsin (log Thresh K x B)