You are on page 1of 219

LET THERE BE MORE

LIGHT
Eugene V. Bobukh

To Make This Presentation


( )

About 70 hours were forever lost for


Facebook
6.2 Gb of disk space was occupied
800+ photos were taken
3 4 5 drinking parties were missed!
So Im glad that youve made it here

Disclaimer

( )

This is not a formal photography


course
I am not a professional

I will not cover many very important


things

So I can do it just however it pleases me

<If you dont like my style, hire a pro for


$>

Such as laws of composition or studio


setup

I am just an ex-physicist whos done


some 300,000 photos in IR, visual, UV,
VUV and XRAY


But Bugs Are Still Possible
If you spot em, report em!

AND NOW WHEN


EVERYBODYS READY
,

The Very First Photo

~1826

"View from the


Window at Le
Gras by Joseph
Nicphore Nipce

Taken on bitumen
as a media

8 hours exposure

Clearly, there
was not enough
light

Since Then

Photography has gone a long, long way


but there is still often not enough light

Often, We See Something Like That

And It Turns Out Like


That

How Do We Make It Better?

PART 1/5
BASIC THEORY
Practice without a theory is like wine without a toast!

Camera Dissected

Reality is
tough!

We need a
simpler model to
get the essence
of the process

A Spherical Camera in Vacuum

a
b

x
x = 1/(1/F 1/D)
b = a*(x/D)

Usually D >> F, so x F and b a*(F/D)

Exposure (Shutter Speed,


)

Exposure is the time T


between the shutter
was opened and closed
(and the inside of the
camera was
illuminated)
D

Typical T: 1/15th 1/500th of a second (denoted as 15 or


500)
Sometimes: hours to ~ 1/10,000th of a second

Aperture ("")
Light comes from all directions.
photons cross each m2 of a surface near
the camera
How many photons get inside? K = *r2

Bear with
me!
They are
projected on the focus

hemisphere with the area S = 2F2

This is the most complex math of the


whole presentation.

Photons count per m2 of the inner image


is B = K/S = *(r/F)2

F r

Introduce N = F/2r

It is called aperture F-number


The rest will be easier.
I promise
stands
(
) and
behind F in the lens name (e.g., for F/3.5
N = 3.5)

Stay tuned!

Then B = /N2
Lenses with the same N produce images
of the same brightness regardless of the
focal length*!

At F/3.5, They All Are The Same!

OK, only brightness is the same. Almost .

Gates Opened by the Nth of the


Focal Length
D = F/16

D = F/5

D = F/1.4

=> => => => picture brightness => => => =>

The lesser N, the brighter the


image

Bright lenses F/2.8 F/1.2 and less


Most common F/3.5
Dark lenses F/5.6 F/11 and more
Terminology: FASTER == BRIGHTER
Because 2x brighter lens lets you make
2x shorter (faster) exposures and
capture 2x faster motions with the same
blur

ENOUGH THEORY FOR


NOW
A practical problem:
It is too dark.
How do you let more light in?

New Gates Syndrome

How do you let more sheep in?

Two Options

1.

2.

Keep the gates open longer == Increase


Exposure
Open them wider == Reduce Aperture

PART 2/5
EXPOSURE

Increase Shutter Time?


It seems simple
Not enough light at 1/30th of a second?
Make it 1/10th
Whats wrong with that?

Handshake Hands Shake


Focal length
reciprocal rule:
T [sec] < 1/F
[mm]
(In 35 mm
equivalent)
E.g., at 50 mm
exposure should be
faster than 1/50
sec

Image Stabilizer

A Shamefully Expensive Solution


Works by floating either the lens or the
matrix to compensate for small vibrations
Exposure gain: 220x
$$$!
But if you can afford it, go for it!
A must have for lenses > 200 mm

Image Stabilizer Effect


300mm, 1/13 sec, handheld
IS OFF

300mm, 1/13 sec, handheld


IS ON

WHAT ELSE?
Lets understand camera shaking

Modes of Motion in 3D
space

Any complex motion


could be projected
into a sum of six
independent modes:

||

Two translational
One translational ||
Three rotational

A push with the


same momentum
will produce roughly
the same amplitude
motion along either
of the modes

Picture Blur x from =1mm


Move

Translational : x =
*(F/D) 10 m 2
pixels
Notable, but may
be acceptable

r
||
F = 35 mm

xF

D = 3.5 m

Translational ||: x =
*(F/D)*(a/D) 3 m
0.5 pixels
Negligible
Rotational: x =
1 mm = 200 pixels!
=> Trash

Conclusion: rot >> > ||

So The Camera Must Not


Rotate!

Not even by 0.005 millimeters (the size of the sensor pixel)!


How? Attach it to something that is more difficult to rotate!
Tripod
Always grab the camera with two hands
or with three (gently press it against your head)

Get solid support for both elbows

Your body
Better: yoga pose
Better: table

Press the camera against a pole, a tree, a window, or a wall

Better rock full body forth and back rather than let your camera dance
alone

Sometimes even in a car or an airplane. Translational vibrations are more


tolerable than rotational of the same amplitude

Ideal: place it on a solid surface and leave

People shoot GS satellites with that!

LETS SEE HOW IT


WORKS
Same subject
300 mm, F/5.6, 0.5 sec, no Stabilizer
Four shots in each position
2nd best shot from each series is presented

One Hand

Two Hands

Two Hands + Forehead

Two Hands + Forehead & Lean on the


Wall

Yoga (Meditation) Pose

Two Hands, Elbows on the Table

Camera Pressed Against the


Doorframe

Again, this is 300 mm handheld @0.77 sec (yes,


I accidentally switched), no IS, and not even

Further Reduce Hands Shake

Avoid heavy exercising a few minutes prior to a


shot. Give your hands time to relax

Minimize the push button motion

Same about heartbeat


Half-press, relax, gently finish it with one finger motion
For tripod: remote control, timer, exposure delay, or
manual cap removal

Breathe as if shooting a real weapon

Inhale
Aim
Exhale naturally & half-press
Relax & hold for about a second
Gently finish the push motion

Burst Shooting

Shoot 2-7 images in continuous mode


With no motion, no change in pose, no
pauses
The 2nd image is typically better than the
1st

No residual shaking due to button pressing

Sometimes, the 5th is even better

I dont know the exact reason. Wanna


research?

Burst Shooting Sample of


6
BEST SHOT

WORST SHOT

Get a habit of shooting multiple frames at a time.


Two per subject should be your minimal rule.

OTHER EXPOSURE
PROBLEMS?

People Move!

and please dont think that finding a good sample of that badness was easy

What part of the body is the


most important?

, !
Hand? Shoulder? Face!

We even have a special area of the brain


dedicated to parsing it!

And on the face the most important are


eyes!
When eyes are sharp, the whole picture
looks mostly OK even if the rest of the
body is blurred

=> Track eyes with the camera

Face & Eyes Tracking

Face & Eyes Tracking

Dance Blur

Light: you need 1/30 or longer!


Dancers: to catch us, you need 1/160 or
shorter!
Tragedy? Not really
Dance is a more or less random set of periodic
harmonics
At some moments of time, motion is slow
enough to catch it
How often?
Various theoretic estimates exit. Lets find out!

Experiment Description

Set up the light


Set up good music
Dance!
Do *lots* of photos
at random moments
at various exposures
Count how many
have eyes sharp
Finally, say kudos to
Alena Birukova for
key participation

Results
Probability of a good shot:
P = 1/(1+Z), where
Z = 0.06*(T/Tnormal)2.
If normal is 1/160, and you
can do 1/30 only:
Z = 0.06*(160/30) 2 = 1.71,
and good photo chance is:
P = 1/(1+1.71) = 37%
So keep shooting!
In fact, even at 12 times slower
exposure you still get about
10% of the pictures OK(*).

Same (but less controlled) research


conducted with Leavenworth Oktoberfest
dancers in 2010 resulted in a similar power
trend with an exponent of 2.4

===================
==
(*) Subjective bias included

Little Break

Problems E-H
E.

F.

G.

H.

[Essential] Learn how to switch you camera into Shutter Priority mode. What
is the minimum and the maximum exposure your camera permits? Tweak
the exposure while aiming the camera at a well-illuminated subject. How
does the aperture setting change in response? Can you summarize what is
that the camera is trying to do?
In Shutter Priority mode, take images of a running person at 1/20 th, 1/80th,
and 1/320th second (you may need bright light or high ISO for that). Discuss
the differences.
Shoot a static scene in automatic mode. Note the F-number and shutter
speed the camera set. Switch to Manual Mode. Shoot the same scene with
twice longer and twice shorter exposures at the same F-number. Why the
difference?
Set the camera to exposure 2x longer than the focal length reciprocal rule
recommends. Switch off the image stabilizer. Try to produce a handheld
image free from handshake blur using burst shooting and any other
techniques presented. If you succeed, try longer exposures (4x, 8x, 16x)
challenge. Hint: you might want to take several 0.25x (shorter) exposure
shots for a reference on what an ideal blur-free image might look like.

Problems 7-9
7.

8.

9.

[Essential] A P&S camera at certain zoom setting has a focal


length of 20 mm which is a 35-mm equivalent of 100 mm. What is
the longest exposure reasonably free from handshake blur at that
zoom?
You are shooting cars racing with a BW camera that has 4500 by
3000 pixels. A car crosses your field of view in 0.5 seconds. What
the exposure should be if you want motion blur to be completely
undetectable?
You are back to film days. You have three 36 frame rolls and your
newspaper ordered 20 good pictures of Leavenworth dancers. For
Polka, blur motion kicks in at the exposures shorter than 1/100
sec. However, it is so dark in the beergarten that you can shoot
only at 1/25th sec. Should you even try to accomplish the task?

Problem (discussion)
Lake Chelan,
2005.
Film @ISO 400,
F/5.6, 35 mm
focal length
equivalent,
approx. 1/400
sec.
Whats wrong
with this
picture?
Is there a way to
make it better?

Problem (discussion)
Night
thunderstorm
over Caribbean
from the airplane.
ISO 4000, 35 mm,
F/1.8, 1/2 sec,
handheld.
Any guesses how
shake blur was
avoided?

PART 3/5
APERTURE

Remember? Brightness ~ 1/N2

The amount of light is proportional to


the inverse square of the F-number!
This is a serious business
[More serious that DoF ~ N]

So Crank the Lens Open!

F-number and Brightness


Aperture

Brightnes
s
(F/3.5 as
100%)

F/11

0.10

F/8.0

0.19

F/5.6

0.39

F/3.5

1.00

F/2.8

1.56

F/2.0

3.06

F/1.8

3.78

F/1.4

6.25

F/1.2

8.51

F-number and Brightness


Aperture

Brightnes
s
(F/3.5 as
100%)

F/11

0.10

F/8.0

0.19

F/5.6

0.39

F/3.5

1.00

F/2.8

1.56

F/2.0

3.06

F/1.8

3.78

F/1.4

6.25

F/1.2

8.51

F-number and Brightness


Aperture

Brightnes
s
(F/3.5 as
100%)

F/11

0.10

F/8.0

0.19

F/5.6

0.39

F/3.5

1.00

F/2.8

1.56

F/2.0

3.06

F/1.8

3.78

F/1.4

6.25

F/1.2

8.51

F-number and Brightness


Aperture

Brightnes
s
(F/3.5 as
100%)

F/11

0.10

F/8.0

0.19

F/5.6

0.39

F/3.5

1.00

F/2.8

1.56

F/2.0

3.06

F/1.8

3.78

F/1.4

6.25

F/1.2

8.51

F-number and Brightness


Aperture

Brightnes
s
(F/3.5 as
100%)

F/11

0.10

F/8.0

0.19

F/5.6

0.39

F/3.5

1.00

F/2.8

1.56

F/2.0

3.06

F/1.8

3.78

F/1.4

6.25

F/1.2

8.51

F-number and Brightness


Aperture

Brightnes
s
(F/3.5 as
100%)

F/11

0.10

F/8.0

0.19

F/5.6

0.39

F/3.5

1.00

F/2.8

1.56

F/2.0

3.06

F/1.8

3.78

F/1.4

6.25

F/1.2

8.51

F-number and Brightness


Aperture

Brightnes
s
(F/3.5 as
100%)

F/11

0.10

F/8.0

0.19

F/5.6

0.39

F/3.5

1.00

F/2.8

1.56

F/2.0

3.06

F/1.8

3.78

F/1.4

6.25

F/1.2

8.51

F-number and Brightness


Aperture

Brightnes
s
(F/3.5 as
100%)

F/11

0.10

F/8.0

0.19

F/5.6

0.39

F/3.5

1.00

F/2.8

1.56

F/2.0

3.06

F/1.8

3.78

F/1.4

6.25

F/1.2

8.51

F-number and Brightness


Aperture

Brightnes
s
(F/3.5 as
100%)

F/11

0.10

F/8.0

0.19

F/5.6

0.39

F/3.5

1.00

F/2.8

1.56

F/2.0

3.06

F/1.8

3.78

F/1.4

6.25

F/1.2

8.51

F-number and Brightness

(Simulated by exposure change; DoF is not actual)

Aperture

Brightnes
s
(F/3.5 as
100%)

F/11

0.10

F/8.0

0.19

F/5.6

0.39

F/3.5

1.00

F/2.8

1.56

F/2.0

3.06

F/1.8

3.78

F/1.4

6.25

F/1.2

8.51

So, Open Aperture Get a Fast Lens

Good news: F/1.8 are reasonably


inexpensive
Canon 50mm $110
Nikon 50mm f/1.8D AF $135

If you can afford any DSLR, you most


likely can afford an F/1.8 lens as well

BAD NEWS?

1. Bye-bye Zoom Lenses

Virtually no zooms faster that F/2.8

Technical reasons

Use Primes
and your feet!

2. $Cost

Pentax SMCP-A 50mm f/2.0


$60
Canon EF 50mm f/1.8 II $110
Nikon 50mm f/1.8D AF $135
Canon EF 50mm f/1.4 USM
$350
Canon EF 50mm f/1.2 L USM
$1440
Canon EF 50mm f/1.0L USM
$2500 (out of production)
Leica Noctilux-M 50 mm
f/0.95 ASPH $11,000
Carl Zeiss Planar 50mm f/0.7:
between $117K and $16M

The lens was designed and made specifically for the NASA Apollo lunar program to capture the far side of
the moon in 1966. <> Stanley Kubrick used these lenses when shooting his film "Barry Lyndon", which
allowed him to shoot the scene only by candlelight. <> In total there were only 10 lenses made. One was
kept by Carl Zeiss, six were sold to NASA, and three were sold to Stanley Kubrick.
In 2011 sold at WestLicht's 19th camera auction for $117K of 2013-$ equivalent

Why Cost ?
3. Anti-reflection more difficult on wider
glass
2. Economy of scale only few people
want such lenses
1. Aberrations! (Or rather the lack of
thereof)

3. Aberrations?
what happens if you do save on cost

Think imperfect focusing


Tough to deal with
Most common classes off the Wikipedia

Spherical Aberration

Light rays near the center and at the edge focus


differently
Image is blurred even when in focus

Chromatic Aberration

Light rays of different color focus differently


Image is color-fringed

Coma

Only paraxial rays focus well


Spots at the edges of the image grow tails

And many, many more

The wider the lens, the more light has to


bend
The more light bends, the more
aberrations
Combating them puts increasingly tough
demand on design complexity,
manufacturing precision and glass quality
as lenses get wider
F/N =>

$$!

$$

Wheres the Limit?

Consumer lenses end at


around F/1.0
Yes there are lenses faster
than F/1.0
Mostly special
applications
So if you use them you
probably work in areas so
special that you should be
giving this talk, not me

,

!

Is there a theoretical
lower limit for the Fnumber?

Introducing Diffraction

F/32

F/8

Any parallel light becomes


divergent after passing
trough a hole
The angle is 1.2/d
Light spot size = F
1.2N
Normally kicks in at closed
apertures
For daylight, 0.5 m,
sensor elements size 6
m, so diffraction is
noticeable at N > 6/0.5 =
12 as some gentle blur

Physics Frowns at N < 0.8

Remember, spot size F = 1.2N

Quantum mechanics: light spot


cannot be narrower than photons
wavelength!

Therefore, a simple lens with N < 0.8


will not be able to completely focus,
ever

In practice this may not matter until


sensor elements are small enough to
detect imperfections

For complex lens, the theory is more


complex so as the result

Still this is the approximate limit

Examples

Carl Zeiss Planar 50mm f/0.7

Tokyo Kogaku Similar 50mm f/0.7

The NASA lens circa 1960s, designed much earlier


A few of these made for Japanese Army in 1944

Signal Corps Engineering 33mm f/0.6


Soviet GOI CV 20mm f/0.5 Mirror lens (-237) (1948)
American Optical 81mm f/0.38 Solid Schmidt Mirror lens
(1967)
And Zeiss Super-Q-Gigantar f/0.33 is a joke (non-functional)

4. Shallow Depth Of Field (DoF)


24 mm, F/1.4

24 mm, F/16

DoF == how much scene before and after the focus point remains sharp
Often, it could be so little that I have to close my 50 mm F/1.4 to F/2

Depth Of Field Facts


<more in Backups>

d N*s*(D/F)2 (follows from school physics)

N is the F-number (aperture setting)


s is the spot size the minimal size of the light spot your matrix
or film can resolve
5-10 m for contemporary DSLRs
3-5 m for P&S cameras

The more N, the more DoF


The more F, the less is perceived DoF

D is subject distance
F is focal length

100 mm F/5.6 has the same DoF as 50 mm F/1.4


Thats one of the reasons youd prefer longer lenses for portraits

A 3.5 m away scene shot in B&W on 1.5 CF camera with


3500 pixels wide sensor at F/3.5 will have a DoF of 35
centimeters!

Will A Larger Matrix Have Lesser


DoF?

No
The image does not
change because I shoot
only on the central part of
the matrix!
It may if you also re-fit it
to the screen, or change
matrix element size, or
scale the lens
proportionally, etc.

SO I PROMISED NO
EXPENSIVE UPGRADES
Maybe F/1.8 meets that definition.
But there is still not enough light.
Whats next?

The Next is a Short Break

Problems A-D
A.

[Essential] Learn to switch your camera to aperture


priority mode and to fully manual mode. Tweak aperture
settings in each of the modes. What are the maximum
and the minimum F-number your camera permits? Which
would produce a brighter image?

B.

In fully manual mode, make a well-exposed shot of a


subject of your choice at F/8.0. Repeat the shot at the
most open and most closed settings. Discuss the
difference.

C.

In aperture priority mode, make three shots of a brightly


illuminated subject a few feet away at minimal,
intermediate, and maximum F-number. What differences
can you see?

Problems 1-3
1.

2.

3.

[Essential] An image is taken with a 50 mm lens, F/5.6,


1/200 sec. The lens is replaced with a 35 mm one. At the
same exposure, what the aperture setting should be to
produce an image of the primary subject with the same
brightness?
An image is taken at F/8.0, 1/400 sec. It is blurred from
subject motion. To be sharp, it needs to be taken at
1/1600 sec. What the aperture setting should be to
preserve the same brightness?
At 35 mm, F/3.5 the depth of field on the portrait was 35
cm. The lens is replaced with 70 mm, F/1.4. If shot from
the same distance, what the depth of field would be?

Problem
ISO 3200, F/1.4,
50mm, 1/125 sec,
subject distance
7.1 m. Effective
sensor element
size 8 m.
What problems
can you identify
on the image?
Can those be
fixed by adjusting
aperture settings?
What is the
calculated DoF
here?

Problem

Left: 48 mm, F/32, 1/25 sec. Right: 48 mm, F/5.3, 1/800 sec. Why the difference

PART 4/5
THE BATTLE OF SENSITIVITY
A Damn Bloody One

What is ISO rating?

A measure of sensitivity to light


Applies to film, digital sensors, and all other lightcapturing media
The more ISO, the better sensitivity
The relation is linear

E.g., a photo at ISO 400 would need twice exposure


time as the same photo at ISO 800

Typical films ISO is 50 1600


Typical digital sensors ISO is 100-200
But then the camera amplifies the result to turn it
into values sometimes as high as 3200-204800

How Much Is One ISO?

It Is Somewhat
Complicated

Because the definition ties it to


humans eye subjective
perception

So, after simplifying and


neglecting many things
Take 1 square meter of black-andwhite film rated ISO 1
Illuminate it with green light @555
nm
Wait until it absorbs 0.00117
Joules of energy
Process
Discover that it is 1.26 times
darker than unexposed film

And it is non-trivial

On the question of how historical reasons


lead to all kinds of crazy figures, see the
horse ass vs. Space Shuttle account.

For ISO X all the same except X


times less energy

If You Think This Was


Complicated

consider digital, where things are way more obscure

++Color
Five different techniques are officially defined by ISO
standard
Strong marketing distortions
JPG & gamma conversion
The proper level of exposure is somewhat subjective

So, with some further oversimplifications


Digital sensor rated at 1 ISO requires about 0.0125
Joules of absorbed green light energy per square meter
to be properly exposed (details in backup)

The energy that *enters* the camera may be much greater


due to the color filter in front of the sensor

To Summarize

The more ISO, the brighter the image


Stop worrying and love high ISO

Too dark? Boost the ISO!

At least 1600
Essentially, Photoshop does the same

What can possibly go wrong?

NOISE!

What Does It Look Like?

The Higher ISO, The More Noise

Noise is complex
The whole Big Science there
Multiple sources
We will consider those we can practically
deal with
Expect to meet many ugly noise
examples prepared with care and on
purpose

Discretization Noise

Arises because digital world has only


small number of discrete values (e.g.,
256 for JPG) while Mother Nature has way
more
Manifests as
Sharp color transitions
Poor JPG quality in dark areas

Discretization Noise

Severely (at least 30x) underexposed scene shot in RAW and boosted digitally

Discretization Noise Mitigations

Digitize as late as possible


Right Exposure First, Photoshop Last
Not the other way around!

Discretization Noise Mitigations

Use RAW/NEF
It has 12-14 bits per color channel >> 8
for JPG
14 bits work really better for dark areas
boosting and JPG conversion math

Especially since you can run it on a PC


processor with more options and power

RAW/NEF PRIMER

50%-exposed JPG (ISO


6400)

boosted 2x in Photoshop

Same Boost & NR from NEF

8x Underexposure Example
ORIGINAL JPG

FROM NEF

But even 12 bits wont help much if you are more than ~4x underexposed

RAW Benefits are Overrated

There is really no need to shoot just


*everything* in RAW as some recommend
RAW really helps if you are 1.25x4x
under the target
Less is correctable in JPG
More not correctable even in RAW
Overexposure? Bracket! 3 JPGs 1 NEF

Other downsides of RAW


Takes significantly longer to work with on the PC
Occupies storage

RAW vs. JPG


When RAW is preferable

High ISO near the limits of your


camera
Subject brightness fluctuates >2x
and faster than you can bracket
Light color is weird or fluctuates
wildly yet youd need to balance it
You have a chance for 1-2 shots only
and cant predict the light
The scene dynamic range is well
over 100x
You are shooting for scientific
quantitative measurements
Your image will be subject to color
profile transformations (e.g. for
printing)

Consider JPG if

Light is strong, nearly constant,


with well-defined color
temperature (e.g., Sun, tungsten)
Or bracketing is sufficient to
curtail light uncertainty/variation
The scene is soft, with no harsh
illumination variations
You shoot 100s or 1000s frames
The shots are of low value

Your time is a non-renewable asset!

No significant post-processing is
expected

Electronic Noise

A whole zoo of phenomena driven by


imperfection of electronics
Diodes produce unrelated dark current
Capacitors leak
Resistors make thermal noise
Pixels are not made equal and vary in sensitivity
Read errors & overflows
Accumulation and amplification errors
and more

Manifests as salt, bands, and dead


pixels

Electronic Noise Example 1

See colored dots, whitish salt?

Electronic Noise Example 2 (Band


Noise)

ELECTRONIC NOISE
MITIGATIONS

1. Get a Good Camera

$
$$?
$$$ #%^&@!!!

2. Find the Sweet Exposure

Too short its dark


Too much its peppered
Experiment!

3. Dark Frame Subtraction

Helps with long exposures over 1-10 sec


Do enable it!
Called Nigh ISO NR/<Noise Reduction>
on most cameras

Dark Frame Subtraction Effect


OFF
ON

ISO 6400, 20 sec

4. Ice Pack

OK, I dont mean that literally


But most electronic noises are strongly
temperature-driven
So keep the camera as cool as possible
Plan for tough dark frames first before
camera gets hot from the rest of the work

To Prove The Point


Please Dont Try This At Home

1.
2.
3.
4.
5.
6.

Take a high ISO photo


Get an ice pack
Apply it to the camera
Wait 20 minutes
Repeat the photo
Compare!

Ice Pack Effect


Before
After

ISO 6400, 15 sec, all NR off

AND FINALLY,
THE MASTER EVIL
Quantum (Shot) Noise

Quantum Noise Origin

At ISO 1, it takes about 0.0125 Joules (1.5x) of energy


per sq. meter of a sensor to get the right exposure with
green light @555 nm
At ISO 1600, that is 0.0125/1600 = 7.8*10 -6 Joules per m2
24x18 mm sensor has 4000x3000 elements (12 Mpx). So
one sensor element is effectively 6x6 m
Therefore, it takes 7.8*10-6 *(6*10-6)2 = 2.81*10-16 Joules
to get one element well-exposed
One photon of green light carries the energy E = hc/ =
3.584*10-19 Joules
So, one sensor element typically gets K=
2.81*1016/3.584* 1019 = 785 photons at proper exposure
@ ISO 1600

Quantum (Shot) Noise


The key word is typically
because it is subject to Quantum Mechanics rules
where everything fluctuates!

Its just like rain droplets

And a variation is about K = K


If an average photon count is 785, quantum fluctuations are
about 785 = 28 photons

Even on an *absolutely* uniform image, some sensors will get 761


photons, some -- 800, or 777, or 809, etc.
So it will not look uniform anymore => noise!

And that is a freaking 28/785=3.6% of a variation!


In most cameras, this is the prevailing source of noise at ISO
>~1600

Roughly independent of the wavelength in 400800 nm range where


quantum efficiency response of a typical sensor is approximately flat

WHAT DOES THE


ENEMY LOOK LIKE?

ISO 200, noise 1.3%

ISO 6400, noise 7.1%

Repeat. Notice Fluctuations?

Quantum (Shot) Noise

It is *completely* random
Einstein believed that God doesn't play dice with
the world
Even if he was right, as of 2013 quantum
fluctuations remain (both theoretically and
practically) the most random and unpredictable
phenomena we know

Best RNG, ever!

No digital filter, ever, can separate it from


information (in a general case). It is
randomness in its most pristine and purest
form.

WHAT DO WE DO ABOUT IT?


Some Mitigations Techniques

Quantum Noise: Exposure

Get a good exposure


Quantum noise/signal ~ 1/(signal)
Better properly expose first rather than
boost both signal and noise in Photoshop
later

Quantum Noise: Larger Pixels

The more photons are represented by


each screen pixel, the better noise
averages out
This concept covers a large group of
mitigations from camera choice to
shooting black-and-white

PART 4.1
LARGE PIXELS
VS.
QUANTUM NOISE
Let me know if we need a short
break

Sensor Element Size

Sensor (mm)

Type

Aspect
Ratio

1/3.6"

4:3

7.056

5.000

4.000

3.000

1/3.2"

4:3

7.938

5.680

4.536

3.416

1/3"

4:3

8.467

6.000

4.800

3.600

1/2.7"

4:3

9.407

6.592

5.270

3.960

1/2"

4:3

12.700

8.000

6.400

4.800

1/1.8"

4:3

14.111

8.933

7.176

5.319

2/3"
1"
4/3"

4:3
4:3
4:3

16.933
25.400
33.867

11.000
16.000
22.500

8.800
12.800
18.000

6.600
9.600
13.500

APS-C

3:2

n/a

30.100

25.100

16.700

3:2

n/a

43.300

36.000

24.000

4:3

n/a

69.700

56.000

41.500

35
mm
645

The larger, the better

Dia.
Diagonal
(mm)

Width

Height

Look up sensor width


Element Size Sensor
Size / Horizontal Resolution
Quantum noise 1/
(Element Size)
Typically Full Frame DSLR <
1/1.5 frame < P&S < Phone

The table is from http://


www.dpreview.com/news/2002/1
0/07/sensorsizes

Insane marketing. 20 Mpx on a


1/2.3 sensor? Its 2 m! At
ISO 12800 fluctuation/signal is
30%!

Between the same frame


sizes, it may be better to
choose a camera with
fewer megapixels!

Lower Resolution

In most cases, you dont need 20 Mpx


Facebook photos are 2048x1368 = 2.8 Mpx
Most Web pictures are barely 1 Mpx
12x8 prints @ 75 dpi in color = 2.2 Mpx
WQXGA monitor is 4 Mpx
Apple 15 MacBook Pro's Retina Display is 5 Mpx

If 12 Mpx ISO 5000 pic looks noisy @100%, it may


still be OK when converted to 3 Mpx for Facebook
The benefit is purely due to scaling when viewing
There is (almost) no difference between setting your
camera to lower resolution or reducing a large picture
obtained from it in Photoshop later.
But at least you save memory and processing time

Resolution Comparison
12 Megapixels

3 Megapixels

Again, keep in mind that the improvement is purely due to


scaling. If viewed at the same feature size, there will be
(virtually) no difference

Digital Filters

that trade resolution for smoothness


Photoshop legitimate

Reduce Noise, Blur, Smart Blur, Gaussian


Blur and alike essentially average your picture.
They lose information but gain smoothness.

Photoshop illegitimate

Creative & Artistic filters (various Brush


Strokes, Film Grain) spoil more than just
noise but do it in a subjectively appealing
manner so you can gain another 2-10x of a
missed exposure!

Filters Primer: Original

Filters Primer: Reduce Noise

Filters Primer: Angled Brush

Angled Brush & Female Models

Camera JPG Settings


A camera-dependent game of modest
(1.2x-2x) benefits. For Nikon D90:

Output mode: Portrait vs. Standard vs. Vivid


Saturation mildly reduces noise
Contrast drops noise but loses shadows
Sharpening drops noise & resolution

JPG Portrait Default

JPG Portrait -2 Sh, -2 Sat, +2


Con

JPG Noise Reduction


Off vs. Low Vs. Normal vs. High

About 2x gain for D90

Yes, you will further lose some sharpness


Sharpness is way overrated. You really want the
whole world to see skin imperfections of that
girl?

JPG Noise Reduction Low (ISO


6400)

JPG Noise Reduction High (ISO


6400)

Black & White

If things are really bad


and you still need extra
1.5x exposure...
Sacrifice color!

After all, white is a color


so as black

Why would that work?

I thought I knew.
But testing confirmed my
predictions wrong.
So I dont really know for
sure. I just see the effect.

In Color

BW

Yes, It Does Depend On Color

in a more complex ways than my original theory


predicted.
So as Simple Demo grew into Lab Work then into
Research Project I decided to stop.

A Note on Strongly Colored


Light
1.

Forget about white balance


1.

2.

There is no @%%ing white possible when


all you have is orange + UV

Underexpose by 0.5 3 stops


1.

Otherwise, your camera will try to collect 1x


light via 0.25 channels. What happens? It
oversaturates.

White Balance And Noise

Noise loves underexposed channels


So lets keep color balance as close to
white as feasible
Not much gain but sometimes even 1.25x
saves the shot

White Balance: Too Cold

White Balance: (Almost) Normal

Problems I-L
[Essential] Locate ISO sensitivity settings in your camera. What are the
maximum and minimum supported values? By shooting the same subject, find
the maximum level of ISO that produces tolerable level of noise (per your
discretion).

I.

Locate the setting to shoot RAW+JPG (at the same time) in your camera. Set
ISO to the maximum. Shoot something. Compare image quality in RAW and in
JPG.

J.

Locate all of the following settings in your camera and test their effect on JPG
image quality:

K.

I.
J.
K.
L.
M.

L.

JPG output mode and conversion settings.


Noise suppression/correction.
Black-and-white mode.
White balance adjustments.
Image size.

[Challenge]: By playing with these settings, can you get a picture with tolerable
noise at ISO one step higher than derived in Problem I?

Problems 10-11
Camera

Pixel count

Sensor size

8 million

4.5 x 3.4 mm

Canon PowerShot
A200

2.1 million

4.5 x 3.4 mm

Minolta DiMAGE X

2.1 million

5.3 x 4.0 mm

5.2 million

7.2 x 5.3 mm

Samsung Galaxy S3

Olympus C-5050
Zoom
Sony DSC-F717
Canon EOS-D30
Nikon D1
Nikon D100
Nikon D90
Canon EOS-1Ds

5.2
3.2
2.7
6.3
12..2
11.4

million
million
million
million
million
million

8.8 x 6.6
22.7 x 15.1
23.7 x 15.6
23.7 x 15.6
23.7 x 15.6
36 x 24

10.

mm
mm
mm
mm
mm
mm

11.

Arrange the cameras in the


table in the order of their
expected quantum noise at
fixed ISO. Use http://
www.dpreview.com/news/20
02/10/07/sensorsizes
for help if needed. Assume
sensors occupy 100% of
cameras matrix area.
If Nikon D90 quantum
noise is 3.6% at ISO 1600,
what would be quantum
noise for other cameras in
the list?

Problem

Of the images on the next page, which


would you consider better shooting in RAW,
JPG, or either? Why? What factors may
change that decision?

<BR>

PART 5/5
STANDARD TOOLS & MISC.
In a rather random order

What is the most important


property of a tripod?

Sturdiness and
bulkiness!

Rotations,
remember?

Turn Image
Stabilizer off

Or it may try to
stabilize the
whole tripod

Built-In Flash

Every programming language has its uses. For


every problem, there is the best programming
language. But avoid COBOL if ever possible
Photographers often despise built-in flash
Downsides?
Strong sharp shadows along the view
Kills depth
Sharp reflections

But in fact, it is still a very useful tool


Always with you
Small and compact

So how do we make the best of it?

BUILT-IN FLASH CASE STUDY

Built-In Flash
Step 1: No flash
Awful shadows.
I look 50 and as
if been
constantly
drinking for the
past 29 years

Built-In Flash
Step 2: flash ON
Shadows gone.
Depth, too. I
look flat.

Built-In Flash
Step 3: Switch to
manual. Adjust flash to
shoot at 60% power
(yes, you can). 50%
light should come from
flash and 50%from
ambient illumination.
Shadows filled in, the
look is more natural.
But why is it all red?
Ahh, white balance is
set for Auto.

Built-In Flash
Step 4: Set
White Balance
for
Incandescent.
Brr
Maybe set it for
flash?

Built-In Flash
Step 5: Set it for flash
Now its all yellow
again. Hmm
Rooms light is
yellow. Flash is white.
What is the correct
white balance here?
None! Compromise or
Photoshop at best

Built-In Flash
Step 6: Think.
What if flash
was yellow, too?

Built-In Flash

Built-In Flash
Step 7: Test it.
Enjoy!

And that does not have to be a napkin

In clubs and on parties,


there are plenty of
wonderful opportunities
You can use it to control
light direction, too.
Virtually any colored
plastic film will do as well

Eye vs. Camera Spectral


Sensitivity Differences

Unprotected matrix sees near IR


Protected is closer to human eye, but at
far blue (~400 nm) is often more sensitive

Way to go when you need to create darkness


for humans but bright light for your camera

Digital is a poor media for deep violet


(would tend to express it as either purple
or blue).
Use film + projector if that is critical.
UV sensitivity drops may still want to use film
for special applicaitons

Light Through A
Viewfinder
I was at Mauna Kea. The view
on the left was OK when I
was taking it, but completely
dark when I was moving to
the front of the camera. Any
hints why?
Yep. The sun was shining
through a viewfinder and
skewing the exposure some
10x times.
For night shots, a street light
in the back can be a killer

Branching Question

Do I need to talk a bit about film?

Film vs. Digital

Where film is still better than digital?


Sensitivity no
Resolution no except for special areas
like holography (up to 5000 lines per
mm)
Noise no
Color! But only if you dont digitize

Reciprocity Failure

Also known as
Film exposures longer than 1 sec deviate from
proportionality to the amount of light
For a scene, the exposure is 1 sec. Now the amount of
light drops 16 times. What the new exposure should be?
Simple theory: 1*16 = 16 sec
Reality: it will take even longer by a factor of approx T
times. Thats it by (16) = 2x, or 32 seconds in the end

Grows into a problem when you do multi-minute


exposures

1 hour (in theory) turns into 8 hours == whole night

Digital does not have that problem

Film Push-Processing

Lets you gain


up to 2-4
times at the
cost of higher
contrast and
more noise

ISO 1600
picture
processed as
6400 (which
was kinda cool
back in 2005!)

STANDARD EXPOSURES

Is There One?

Night == High Dynamic Range


The difference between the brightest and
the darkest parts of the scene can be
millions times
What do you really want to see?

You Can Even Turn Night Into


Day

As I foolishly did once near Hells Canyon, Oregon.


This is night. That shiny blob is the Moon Total Failure

There Is No Right
Exposure

You choose whats right

Reality is more malleable at night

But a different question is legitimate:


If I shoot a snow under the Moon, what
should the exposure be to make the
picture 18% gray?
Welcome the Table of Standard Exposures

Table of Standard
Exposures

Crude guide only, 3x easily.


Experiment!
Standard means ISO 400, F/3.5, ideal
lens

You know now how to adjust for different


settings

White diffuse surface (snow, paper)


Light sources shine perpendicularly (the
Sun is right above, the light falls
normally, etc.)
Point sources (light bulbs, fire) are 1

Standard Exposures, sec

Sunlit landscape: 1/30000


Cloudy landscape: 1/500
Sun at the sunset: 1/200

2 minutes after sunset: 1/10


10 minutes after sunset: 2

Landscape under full moon: 10


Landscape under half moon: 150
Starlit landscape: 6000
Incandescent 100W: 1/60
Incandescent 40W: 1/15
Fluorescent 40W: 1/80
Smartphone screen: 1/2
Typical campfire: 1/21/10
Candle or match fire: 1-3
Inside a night club: 1-10
Cigarette: ~1000
Enough light to read at hands length: 8
Enough light to see colors: 100
Enough light to see another human after adaptation: 10 6

A Field Algorithm, 1/2


1.
2.

Sync the clock


Switch to Manual Mode that will surface all your errors
1.

3.

4.
5.

Utilize any light you can find that does not kill the scene. Consider flashlights, lighters, or
even built-in flash (but remember to adjust and color it)
Set the aperture to the maximum, except if you need more DoF or hit aberrations
Set the shutter speed (exposure)
1.
2.
3.

6.
7.
8.

9.

If you know what you are doing, use A or S modes

Handheld: start with 1/F rule, increase later if needed


Moving people, dancers: start with blur threshold (around 1/100 by default)
Tripod: per your patience

Adjust white balance (if needed and possible)


If you have Image Stabilizer, adjust its settings
Make sure your Noise Reduction is ON. If youre shooting JPG and understand its
conversion settings, adjust them to better combat noise. If your subject are people,
Portrait output mode for Nikon is recommended
Set the minimal ISO that exposes at least 2/3 of the histogram width for JPG [1/3+ for
RAW], or has the right brightness in those parts of the scene you care about
1.
2.
3.
4.

If you dont know how to use histograms to check the exposure, ask. This is important.
Fear no evil. If you need ISO 12800 so be it.
If too noisy, consider RAW, lower ISO, boost noise reduction, B&W, or be ready to Photoshop later.
If still not enough, go back to step 3 and reiterate. Sacrifice DoF, shutter speed, resolution, or color. Change
light.

A Field Algorithm, 2/2


10.

Shoot

1.
2.
3.
4.

Hold the camera with both hands. Be stable. Breathe right. Try to find firm support for elbows, lean against the wall/tree
or press the camera against it
Shoot 2-7 frames per scene in burst mode
If not sure about exposure, or getting off target, switch to RAW. Or, bracket in JPG in 3 shots for <-0.7, 0, +0.7>.
Manual focusing with lenses faster than F/2.0 is nearly impossible in poor light. Use auto-focus. To aid it, briefly
illuminate the subjects with additional light (or built-in focus assist illuminator) before shooing if thats appropriate. Set
focusing to the Center mode, aim the center square at high contrast transitions, then hold, recompose and finish the
shot.

11. Never

trust yourself. Bracket. Constantly tweak settings (especially exposure). Experiment. Break a
few rules If nothing else, at least youll gain the experience
12. Check results frequently and adjust on the go
13. Diversify you results. 50% of success is uniqueness. Here are some ideas to try:
Zoom (if your lens can)
2. Move closer or further away. Shift to the left or to the right. Shoot from the knees level. Shoot from the above. Shoot
from under the chair, from the side, from the back.
3. Change foreground, add some objects to it. Change the background. Frame your main subject with some foreground
objects (doors, passages) this makes a cool effect.
4. Move your light around if you can
5. Ask your models to move, pose, or regroup
6. Play with colors, white balance, light sources
7. Go somewhere else and shoot something else
8. Talk to other people to exchange lenses, filters, or even cameras
9. Look for contrasts in ideas, impressions, colors
10. Look for unique points of view or compositions
1.

14.

Try to get at least 100 technically correct photos (without duplicates) per assigment

ALMOST THE END

Starring in Episodes

Vlad Korolyov
Yvgeny Epshtein
Eugene Bobukh
Dima B.
Andrey Ulanov
Ylyana Shershunova
Alena Birukova
Natalie Kachook
Ksenia Du Van
Konstantin Roslyakov
Yulia Roslyakova
Rodion Degtyar
Michael Tverskoy
Sergey Khristoforov
Maria Konstantinova
Kirill Gil
Alina Gil
Michael M Grinshtat
Sergey Prozhogin
Svetlana Prozhogin
Ivan Sydorenko
Olesya Voronina
Michael Pedchenko
..and many more

Our Gains
Switching from F/3.5 to F/1.8: 4x
or F/5.6 to F/1.4 16x
Holding the camera right: 2-10x
Burst Shooting: 2-5x
or knowing dancers blur spectra: ~10x
JPG -> RAW/NEF: 4x
or camera JPG settings: 1.52x
Keeping the camera cool:1...1.5x
Photoshop filters & lower resolution: 2...10x
Right white balance or color drop1...1.5x
===================================
Total: 1x48x32,000x

THE END
Questions?

BBB
BACKUP, BONUS, BONEYARD

F-number and Brightness

(Simulated by exposure change; DoF is not actual)

Aperture

Brightnes
s
(F/3.5 as
100%)

F/11

0.10

F/8.0

0.19

F/5.6

0.39

F/3.5

1.00

F/2.8

1.56

F/2.0

3.06

F/1.8

3.78

F/1.4

6.25

F/1.2

8.51

F/0.7

25.0

Why? Compare The Shadows!


F/3.5

F/0.7

Why larger matrices will not


have lesser DoF?

Thought experiment of taking a picture and


cutting away the edges of the matrix
However.

When viewed on the screen, the perceived DoF


may change
The *actual* information in the picture will not

If all sizes are increased N times (lens, camera,


sensor element to preserve megapixel count) it
will change 1/N times
Thats why P&S will have lesser DoF

If sensor element size wont change, DoF will


change as 1/N2.

DoF Formula is Simple


Practical Applications May Be Confusing

300 mm, F/5.6

50 mm, F/5.6

Generally, d drops as
1/F2
85mm F/8 will have
the same Depth Of
Field as 35mm F/1.4!
but if subject
occupies the same
part of the image, d
does not depend on F!
but, d/D
(perceived DoF) still
drops as 1/F
Thats why youd
prefer 85 mm over 24
mm for portraits

What about F4?


1.
2.

3.
4.

5.

6.

DoF N/F2
If I want to keep fixed
DoF, I must maintain N
F2
Amount of light K 1/N2
For fixed visible
impression, exposition T
1/L
Therefore, T N2 F4 for
fixed DoF
Thats one of the reasons
its so hard to deal with
long lenses

Face & Eyes Tracking

ISO Quantification 1

Film ISO definition:

S[ensitivity] = 0.8/H
H exposure in lux*sec where log10 optical density is 0.1 above fog level, or 10^0.1 = 1.26x
darker
1 lux*sec = 1 lm*sec/m^2
Brightness in lumens = K*P,
P is irradiance in energy units, Watts
K is conversion coefficient between energy units and lumens perceived by humans. It is
strongly wavelength dependent.

For a moment, lets figure out the story with green light
where K = 683 lm/Wt
H = 0.8/S (assuming 1 sq. meter) => P = H/K = 0.8/(SK). If
S = 1 [ISO], P = 0.8/K = 0.8/683 = 0.00117 Joules.

ISO Quantification 2

1.

2.

3.

Digital could be approached in several ways


By compatibility with film. Digital and film of the same ISO should produce similar pictures
under similar illumination. Film @ ISO 1 exposed to 0.8 lux*seconds produces optical
density of 0.1 which corresponds to the beginning of its usable linear range. Typical film
would have a dynamic range of 2-3 (lets say 2.5) magnitudes before it saturates. The
middle of it would be 2.5/2 = 1.25 (Wikipedias ISO definition quotes a point of 1.3 but Im
not certain that figure is relevant). So to move from a density of 0.1 to the middle of the
range it would require about 10^(1.25) = 17.78 times more light. Therefore, this method
predicts 0.8*17.78 = 14.23 lux*sec needed for a good exposure for digital ISO 1 sensor
of 1 m2.
By Standard Output Sensitivity ISO definition: Ssos = 10/Hsos, Hsos leads to saturation of
118 in 8-bit pixels (results in 18% grey after 2.2-gamma conversion as sRGB). So in this
case the needed exposure is 10 lux*second.
By Saturation-based definition and film analogy again: S = 78/Hsat, Hsat the largest
exposure that does not lead to clipped output. Results in 13% grey. That is an equivalent of
full exposure for the film. Half-way exposure would be, as we assumed a 2.5 DR for a film,
78/10^1.25 = 4.38 lux*sec.

Since we dont know which method is closer to reality (marketing anyways is almost
free to choose between any of the five methods), lets take a geometric mean of all
three values for a measure of a good exposure for a digital sensor, thats it:
(14.23*10*4.38)^(1/3) = 8.54 lux*sec.
With green light @555 nm the energy equivalent of that would be 8.54/683 =
0.0125 Joules/m2 absorbed.
What is the corresponding photon count? At 555 nm, photon energy is 2.237 eV, or
(@ 1.602*10^-19 J/eV) 3.584e-19 Joules. Per square meter, at the right exposure that
is 0.0125/3.584e-19 = 3.488e16 photons for ISO 1 sensor, 1 m2 of active area.

ISO Quantification 3

The was for a green light. What about other colors or


other light sources? Say for deep red @ 700 nm do we
need to use 2.9 lm/Wt conversion?
The answer is no if we are interested in energy (or
photons) absorbed by the matrix as opposed to entering
the camera.
Yes, a camera may have a response curve similar to
humans eye. But that is due to a filter sitting before the
sensor, blocking 99% of 700 nm light.
The unprotected sensors sensitivity at that wavelength
is still very high. Roughly, it is flat from 400 to 800 nm
( some 1.5x, which we ignore at this level of detail) in
terms of quantum efficiency thats it, in photons count.
Therefore, we still need about 3.488e16 photons to be
absorbed by ISO 1 sensor of 1 m2 for good exposure
at any wavelength in the 400-800 nm range.
For ISO 1600, thats 1600 times less, or some 2.18e13
photons.
Now take a sensor 6x6 micrometers: 2.18e13*(6e-6*6e6) = 785 photons for a good exposure @ ISO 1600.

Source

K, lm/W

Tungsten

12-20

Sun

93

Fire

0.3 2

LED

50-100

Fluorescent

80-100

Green @555
nm

683

Red @700 nm

2.9

Red @650 nm

81

Orange @615
nm

334

Yellow @575
nm

644

Cyan @500nm

238

Blue @470 nm

89

Purple @425
nm

18

ISO Quantification 4 & Noise

785 photons @ ISO 1600 correspond to a noise level of


785 = 28 photons, or 3.6%
Other reference figures:
P&S camera with 2x3 m2 sensor @ ISO 1600: 131 photons, 8.7%
noise
DSLR @ISO 1600, exposure 10% of the normal: 73 photons, 11%
noise
ISO boost to 12800: 98 photons, 10% noise
Drop resolution 12Mpx -> 3 Mps: 3140 photons, 1.8% noise

ISO 640, noise 2.3%

ISO 2000, noise 4.0%

Quantum Noise: Multiple


Exposures

Used mostly by astronomers


Idea: take P >> 1 photos, carefully
average => noise drops P times
Usually requires special software
Requires scenes that dont change
while you are shooting 100 frames

Noise Fixed Tgt Exposure Mode

Light Color
Red
Yellow
Green
Cyan
Blue
UV

Noise
Noise
EV required Level
Mode
-2.67
4.6069
-2.33
4.4051
-2
4.6693
-2.333
4.1094
-2
5.752
-0.33
6.3131

in BW
Noise Reduction
3.1517
1.4552
3.135
1.2701
2.9089
1.7604
2.8801
1.2293
3.6717
2.0803
3.8183
2.4948

White Balance: Too Warm

Filters: Original

Filters: Reduce Noise

Filters: Angled Brush

Smog & Fog


Are friends. Especially in Contour

B/W skin photography

Go Yellow-Orange

A fine balance between skin reflectivity, depth


of light penetration, and sensors noise

Focusing in the darkness

Spot-focusing on contrast borders works


best

Point and reframe

I tend to carry a small laser pointer for


that

Light reflection in lenses

Usually not a problem


But cheap filters, on non-|| light can eat
up to 10-15% light
Off-center images may be darker

Vignetting

~(Cos)4!
May get serious at the edges
Strongly lens-dependent
Manufacturers generally try to combat it

Film Sensibilization

Hydrogen

One of Experimental Setups

My Room

An Egg in UV Light

You might also like