Photonrate
Time
End of exposure End of exposure
IdealRate of incoming light is constant
RealityRate of incoming light is fluctuating
Photonrate
Time
How much fluctuation?
√nDictated by Poisson (pwasõ) Distribution
For n total photons in exposure,
standard deviation =
Photons collected = n + √n
If n = 10,000 photons,Photons collected = 10,000 + √10,000 = 10,000 + 100 photons
10,000 photon
s
9,900photon
s
10,100photon
s
9,950photon
s
√n, so what?Variable brightness = noise!
Noise monsterOne of the photographer’s worst enemies
0 1000 2000 3000 4000 5000 6000 7000 8000 9000 100000%
2%
4%
6%
8%
10%
12%Proportion of fluctuation
sqrt(n)/n
Pro
porti
on o
f Pho
ton
Fluc
tuat
ion
(√(n
)/n)
Photons
0 500 1000 1500 2000 2500 3000 35000%
1%
2%
3%
4%
5%
6%
7%
8%
Canon A570ISCanon A570IS
ISO100
ISO200
ISO400
ISO800
ISO1600
Pro
porti
on o
f Pho
ton
Fluc
tuat
ion
(√(n
)/n)
Photons
Canon A570IS7.1 Megapixels1/2.5” sensor5.76 x 4.29mm (24.7mm2)Density: 3.48 μm2/pixel
0 500 1000 1500 2000 2500 3000 35000%
2%
4%
6%
8%
10%
12%
Canon 5DNikon D300Sony H9
Pro
porti
on o
f Pho
ton
Fluc
tuat
ion
(√(n
)/n)
Photons
90 900 90000%
2%
4%
6%
8%
10%
12%
Canon 5DNikon D300Sony H9
Pro
porti
on o
f Pho
ton
Fluc
tuat
ion
(√(n
)/n)
Photons
ISO100 ISO200 ISO400
ISO800 ISO1600
Canon A570IS7.1 Megapixels1/2.5” sensor5.74 x 4.3mm (24.7mm2)Area: 3.48 μm2/pixel
0 500 1000 1500 2000 2500 3000 35000%1%2%3%4%5%6%7%8%
100200400
800
1600
images: dcresource.com
0 10002000300040005000600070008000-2%
0%
2%
4%
6%
8%
ISO100 ISO200 ISO400
ISO800 ISO1600
Fuji F306.1 Megapixels1/1.7” sensor7.7 x 5.77mm (44.4mm2)Area: 7.27 μm2/pixel
100200400800
1600
images: dcresource.com
0 100002000030000400005000060000-2%
0%
2%
4%
6%
8%
ISO100 ISO200 ISO400
ISO800 ISO1600
Canon 30D8.2 MegapixelsAPS-C sensor22.5 x 15mm (337.5mm2)Area: 41.16 μm2/pixel
1002004008001600
images: dcresource.com
30D
30D
F30
F30
A570IS
A570IS
8.2 Megapixels22.5 x 15mm (337.5mm2)Density: 41.16 μm2/pixel
6.1 Megapixels7.7 x 5.77mm (44.4mm2)Density: 7.27 μm2/pixel
7.1 Megapixels5.76 x 4.29mm (24.7mm2)Density: 3.48 μm2/pixel
ISO100 ISO200 ISO400 ISO800 ISO16000%
1%
2%
3%
4%
5%
6%
7%
8%Canon A570IS
Fuji F30
Canon 30D
Canon 5D
Pro
porti
on o
f Pho
ton
Fluc
tuat
ion
(√(n
)/n)
ISO
More photowells mean less photons per pixel
Making matters worse, circuitry must occupy some space between each photowell – the more photowells, the more space circuitry takes up.
SummaryMore pixels, smaller sensor => less light per pixel => more noiseLess pixels, bigger sensor => more light per pixels => less noise
In theory, the biggest sensor with the least pixels will give us the best image, in terms of noise.
A 1-pixel sensor would be ideal.
With 1 pixel, we’d have low noise but no detail.
Many pixels => High detail, high noiseFew pixels => Low detail, low noise
The “Megapixel Myth”: Detail vs. Noise
Megapixels: Detail vs. Noise
Facebook profile picture: 4x6 studio print at 300dpi:
5x7 studio print at 300dpiStandard VGA TV:
1080p HDTV: Projector Screen:
8.5x11in, 300dpi magazine spread:10x14in, 150dpi full-page spread in Daily Cal:
Giant 20x30in poster print at 150dpi:
0.03 MP2.16 MP3.15 MP0.35 MP2.07 MP1.92 MP8.42 MP3.15 MP13.5 MP
How many pixels do we need?:
If you only look at pictures on the computer, 2-3MPIf you make non-poster-size prints (4x6, 5x7, 8x10), 3-4MP
More pixels beyond this don’t add detail, but contribute to greater noise
Caveat: assuming same sensor technology
Chrominance NoiseVariation in color
Luminance NoiseVariation in brightness
Random NoiseFluctuation in light input (random distribution of photons)
Fixed Pattern Noise-Consistently reproducible noise-Caused by electronics, sensor defects, heat-Manifested as “hot pixels” or luminance anomalies-Since noise is reproducible (non-random), easily fixed by dark frame subtraction
