Sunny 16 rule?

On 2 Sep 2004 10:14:23 -0700, wrote:

jpc wrote:

For the DaveBox oly 8080

Level standard deviation

230 0.59
193 1.14
134 1,59
87 2.12
41 2.29

If you look at raw JPEG data from any camera you'll get the same sort
of thing; the gamma compresses the highs and expands the lows. You
have to remove the effect of gamma and _then_ do your averages. You
shouldn't assume a gamma either: measure it. Image a calibrated
target (see JPS's http://www.pbase.com/image/32751804 for an example),
against a uniform background. Try to arrange for even illumination,
but you'll use the background to digitally level it anyways (this
corrects for optical vignetting as well). You can then put your JPEG
data against a known axis and fit gamma curves. Measure the gamma
with one set of images, and measure the noise from a different set.
Treat channels separately until proven otherwise.

If you have manual control over the exposure, there is an easier way
using flat-fields.

[...] To say that this camera has 'a' single noise number is meaningless.

One doesn't need a single number, one needs a model.


I've done pretty much everthing you've suggested already with jpg
images. Unfortunately, I realize now that a jpg image is not the right
image to used for these measurements

First--here are my assumptions.

!-All electronic noise will create a noise floor that is independent
of illumination.
2-Exposures are short so thermal shot noise can be ignored.
3-Measured noise is the sum of the noise floor and the photon noise
4 Photon noise is equal to the sqrt of the number of photoelectrons in
the sensor well.
5 Raw data from the camera is linear and has a gamma of 1

I have a calibrated transmittance step gage where the optical density
ranges from ~0 to 3 in 21 steps. My camera has 11 contrast settings
and I've taken sets of jpg images thru the step guage. The
illumination was diffuse sunlight reflected off a white box.

With ImageJ I measured the bit levels in the images of the step guage
images plus the noise at that bit level. When I plotted the bit levels
vs the transmittance of the step guage with a log-log scale I created
a family of curves similar to characteristic curves of film with both
a toe and a shoulder.

When I redo the measurements in RAW mode, the new curves should be
linear. They can then be normalized to a gamma of 1. The signal to
noise ratio and the noise values will also be independent of all
artifacts cause by the camera imaging processing software.

Unless you can see some holes in my logic and procedures, so far so
good.

You said in another post in the thread that there are more than one
noise model. I'm assuming you mean an amplitude noise model, a power
noise model and perhaps others I've nerver heard of.

The photon S/N is giiven by the number of photo electrons in the
sensor divided by the sqrt of that number. Is that ratio identical to
the ratio given by corrected bit level as measured by the proceedure
above divided by standard deviations of the noise at that bit level?
Or are there other complications because photon noise is defined using
a different noise model?

If the ratios are identical I could then use the measured data to
determine the sensor well depth and noise floor. Moreover if I can
borrow an incident light meter that reads in physical units I could
determine the illumination. Based on some reasonable assumptions about
the optics, I could then calculate quantum efficencies, ISO rating and
so on.

Be a bit of a project, but if matters work out, anyone could extract
a considerable amount of undocumention information about their camera
with a relatively simple experiment.

jpc

John McWilliams wrote:

Interesting thread, but under what circumstances - today - would one
want or need the sunny 16? Back when meters were atrocious or it was
hugely expensive to buy a good one, understood. Back when one might have
forgotten one's light meter and wanted an approximation, understood.

I've used it when shooting stock car races, most of which happened under
bright summer daylight. I found too many variations in metering between
light- and dark-colored cars, that shots of light cars were coming out
underexposed and darker cars overexposed, and that AE was too easily
fooled by spots of sun glare or other overly-dark or -light items... and
when you're shooting a variety of vehicles going 'round and 'round under
ever-changing surroundings, those AE-fooling items come and go quickly.
In short, I found my Rebel's AE to be pretty much useless.

I tried just working off Sunny-16, and the results were consistently
excellent: 100ISO film (this was with my 35mm Rebel G), f/5.6 (max
aperture for my Canon 75-300), 1/1000s shutter. The Rebel's AF worked
great, BTW.

I metered off an area of mid-grey pavement and got pretty much the same
exposure readings, too...

John McWilliams wrote:

Interesting thread, but under what circumstances - today - would one
want or need the sunny 16? Back when meters were atrocious or it was
hugely expensive to buy a good one, understood. Back when one might have
forgotten one's light meter and wanted an approximation, understood.

I've used it when shooting stock car races, most of which happened under
bright summer daylight. I found too many variations in metering between
light- and dark-colored cars, that shots of light cars were coming out
underexposed and darker cars overexposed, and that AE was too easily
fooled by spots of sun glare or other overly-dark or -light items... and
when you're shooting a variety of vehicles going 'round and 'round under
ever-changing surroundings, those AE-fooling items come and go quickly.
In short, I found my Rebel's AE to be pretty much useless.

I tried just working off Sunny-16, and the results were consistently
excellent: 100ISO film (this was with my 35mm Rebel G), f/5.6 (max
aperture for my Canon 75-300), 1/1000s shutter. The Rebel's AF worked
great, BTW.

I metered off an area of mid-grey pavement and got pretty much the same
exposure readings, too...

In message ,
jpc wrote:

Be a bit of a project, but if matters work out, anyone could extract
a considerable amount of undocumention information about their camera
with a relatively simple experiment.

I must be very naive, because I think this data should be included in
the manual.

They go through the trouble to give you such a nice Sensor and full
manual control of the camera, and then assume that you're a tourist in
dummy mode and don't need to know how the camera responds. Even the 20D
is only going to have a luminance histogram, and that's probably a JPEG
histogram, as well. How hard would it be to just give 3 separate
histograms for the three monochrome channels, including blackpoint (a
good visual reminder of ISO setting) and everything up to 4095? THAT is
what I need to know, not this color-balanced and clipped JPEG nonsense.

Canon is a deaf company, who thinks they know everything about what
people need.
--


John P Sheehy

In message ,
jpc wrote:

Be a bit of a project, but if matters work out, anyone could extract
a considerable amount of undocumention information about their camera
with a relatively simple experiment.

I must be very naive, because I think this data should be included in
the manual.

They go through the trouble to give you such a nice Sensor and full
manual control of the camera, and then assume that you're a tourist in
dummy mode and don't need to know how the camera responds. Even the 20D
is only going to have a luminance histogram, and that's probably a JPEG
histogram, as well. How hard would it be to just give 3 separate
histograms for the three monochrome channels, including blackpoint (a
good visual reminder of ISO setting) and everything up to 4095? THAT is
what I need to know, not this color-balanced and clipped JPEG nonsense.

Canon is a deaf company, who thinks they know everything about what
people need.
--


John P Sheehy

In message ,
jpc wrote:

Be a bit of a project, but if matters work out, anyone could extract
a considerable amount of undocumention information about their camera
with a relatively simple experiment.

I must be very naive, because I think this data should be included in
the manual.

They go through the trouble to give you such a nice Sensor and full
manual control of the camera, and then assume that you're a tourist in
dummy mode and don't need to know how the camera responds. Even the 20D
is only going to have a luminance histogram, and that's probably a JPEG
histogram, as well. How hard would it be to just give 3 separate
histograms for the three monochrome channels, including blackpoint (a
good visual reminder of ISO setting) and everything up to 4095? THAT is
what I need to know, not this color-balanced and clipped JPEG nonsense.

Canon is a deaf company, who thinks they know everything about what
people need.
--


John P Sheehy

In message o8VZc.305989$J06.123771@pd7tw2no,
Matt Ion wrote:

John McWilliams wrote:

Interesting thread, but under what circumstances - today - would one
want or need the sunny 16? Back when meters were atrocious or it was
hugely expensive to buy a good one, understood. Back when one might have
forgotten one's light meter and wanted an approximation, understood.

I've used it when shooting stock car races, most of which happened under
bright summer daylight. I found too many variations in metering between
light- and dark-colored cars, that shots of light cars were coming out
underexposed and darker cars overexposed, and that AE was too easily
fooled by spots of sun glare or other overly-dark or -light items... and
when you're shooting a variety of vehicles going 'round and 'round under
ever-changing surroundings, those AE-fooling items come and go quickly.
In short, I found my Rebel's AE to be pretty much useless.

I tried just working off Sunny-16, and the results were consistently
excellent: 100ISO film (this was with my 35mm Rebel G), f/5.6 (max
aperture for my Canon 75-300), 1/1000s shutter. The Rebel's AF worked
great, BTW.

You are probably under-exposing the images a bit with the sunny f16. I
say "underexposing" in terms of what the sensor sees. It may seem
necessary to under-expose if you're shooting JPEG, or are shooting RAW
and are not using a converter that can extract all the dynamic range
from the RAW file. There are two basic problems:

1) Even in evaluative metering mode, the FOV varies tremendously from
spot to spot as far as sensitivity is concerned. This makes exposure
swing a bit in scenes that have areas of greatly differing illumination.

2) Unlike most other cameras, the 10D and Rebel have no shoulder to the
highlights, and use a pretty straight-forward gamma correction right up
to the white 255,255,255. This clips all channels at 2.2 stops above
what the camera meters as mid grey, but the RAW data has aboout 3.8
stops of red, 3.2 stops of green, and 3.1 stops of blue above mid grey.
The extra is completely discarded in a JPEG. Some of it could have been
squeezed into the top of the JPEG highlights, as many other cameras do.

The 10D/Rebel, IMO, is only a good camera if you shoot RAW. If you
shoot JPEG, you have to under-expose all the time, and that is not a
free lunch. If you under-expose by 1 stop, you lose one stop of bit
depth (more posterization), the noise is twice as tall (relative to the
signal), and so are the JPEG artifacts.
--


John P Sheehy

In message o8VZc.305989$J06.123771@pd7tw2no,
Matt Ion wrote:

John McWilliams wrote:

Interesting thread, but under what circumstances - today - would one
want or need the sunny 16? Back when meters were atrocious or it was
hugely expensive to buy a good one, understood. Back when one might have
forgotten one's light meter and wanted an approximation, understood.

I've used it when shooting stock car races, most of which happened under
bright summer daylight. I found too many variations in metering between
light- and dark-colored cars, that shots of light cars were coming out
underexposed and darker cars overexposed, and that AE was too easily
fooled by spots of sun glare or other overly-dark or -light items... and
when you're shooting a variety of vehicles going 'round and 'round under
ever-changing surroundings, those AE-fooling items come and go quickly.
In short, I found my Rebel's AE to be pretty much useless.

I tried just working off Sunny-16, and the results were consistently
excellent: 100ISO film (this was with my 35mm Rebel G), f/5.6 (max
aperture for my Canon 75-300), 1/1000s shutter. The Rebel's AF worked
great, BTW.

You are probably under-exposing the images a bit with the sunny f16. I
say "underexposing" in terms of what the sensor sees. It may seem
necessary to under-expose if you're shooting JPEG, or are shooting RAW
and are not using a converter that can extract all the dynamic range
from the RAW file. There are two basic problems:

1) Even in evaluative metering mode, the FOV varies tremendously from
spot to spot as far as sensitivity is concerned. This makes exposure
swing a bit in scenes that have areas of greatly differing illumination.

2) Unlike most other cameras, the 10D and Rebel have no shoulder to the
highlights, and use a pretty straight-forward gamma correction right up
to the white 255,255,255. This clips all channels at 2.2 stops above
what the camera meters as mid grey, but the RAW data has aboout 3.8
stops of red, 3.2 stops of green, and 3.1 stops of blue above mid grey.
The extra is completely discarded in a JPEG. Some of it could have been
squeezed into the top of the JPEG highlights, as many other cameras do.

The 10D/Rebel, IMO, is only a good camera if you shoot RAW. If you
shoot JPEG, you have to under-expose all the time, and that is not a
free lunch. If you under-expose by 1 stop, you lose one stop of bit
depth (more posterization), the noise is twice as tall (relative to the
signal), and so are the JPEG artifacts.
--


John P Sheehy

In message o8VZc.305989$J06.123771@pd7tw2no,
Matt Ion wrote:

John McWilliams wrote:

Interesting thread, but under what circumstances - today - would one
want or need the sunny 16? Back when meters were atrocious or it was
hugely expensive to buy a good one, understood. Back when one might have
forgotten one's light meter and wanted an approximation, understood.

I've used it when shooting stock car races, most of which happened under
bright summer daylight. I found too many variations in metering between
light- and dark-colored cars, that shots of light cars were coming out
underexposed and darker cars overexposed, and that AE was too easily
fooled by spots of sun glare or other overly-dark or -light items... and
when you're shooting a variety of vehicles going 'round and 'round under
ever-changing surroundings, those AE-fooling items come and go quickly.
In short, I found my Rebel's AE to be pretty much useless.

I tried just working off Sunny-16, and the results were consistently
excellent: 100ISO film (this was with my 35mm Rebel G), f/5.6 (max
aperture for my Canon 75-300), 1/1000s shutter. The Rebel's AF worked
great, BTW.

You are probably under-exposing the images a bit with the sunny f16. I
say "underexposing" in terms of what the sensor sees. It may seem
necessary to under-expose if you're shooting JPEG, or are shooting RAW
and are not using a converter that can extract all the dynamic range
from the RAW file. There are two basic problems:

1) Even in evaluative metering mode, the FOV varies tremendously from
spot to spot as far as sensitivity is concerned. This makes exposure
swing a bit in scenes that have areas of greatly differing illumination.

2) Unlike most other cameras, the 10D and Rebel have no shoulder to the
highlights, and use a pretty straight-forward gamma correction right up
to the white 255,255,255. This clips all channels at 2.2 stops above
what the camera meters as mid grey, but the RAW data has aboout 3.8
stops of red, 3.2 stops of green, and 3.1 stops of blue above mid grey.
The extra is completely discarded in a JPEG. Some of it could have been
squeezed into the top of the JPEG highlights, as many other cameras do.

The 10D/Rebel, IMO, is only a good camera if you shoot RAW. If you
shoot JPEG, you have to under-expose all the time, and that is not a
free lunch. If you under-expose by 1 stop, you lose one stop of bit
depth (more posterization), the noise is twice as tall (relative to the
signal), and so are the JPEG artifacts.
--


John P Sheehy

wrote:

The 10D/Rebel, IMO, is only a good camera if you shoot RAW. If you
shoot JPEG, you have to under-expose all the time, and that is not a
free lunch. If you under-expose by 1 stop, you lose one stop of bit
depth (more posterization), the noise is twice as tall (relative to the
signal), and so are the JPEG artifacts.

Someone wrote a few days ago that his rule of thumb with that camera
(well, one of 'em!) is to use RAW outdoors, and hi-jpeg inside.

Makes sense to me; any drawbacks?

--
John McWilliams