RAW and ISO settings

His reply has nothing to do with your rude reply to the original poster.
Your reply was still RUDE.

wrote in message
ups.com...
McLeod wrote:

Does anyone understand what are the mechanics behind the ISO settings
for RAW images?

Yes, I understand them. At least for the Canon 1D Mark II: there are
three buttons, a finger wheel, and a thumb wheel. Maybe the Nikon D70s
has the Telepathic Adaptor which can lock onto the mind of the
photographer, thus dispensing with these crude mechanical implements?

I think the poster was asking a more generalized question than you
responded to, but hey, who am I to stop you from making a jerk of
yourself.

You "think"?

Usually the lowest ISO setting of your digital camera is going to be
where you get the best picture with the lowest noise. Higher settings
are amplified, I believe, so noise is increased.

You "believe"?

They are only in increments of 100, 200, 400 to remain reciprocal with
shutter speed and aperture settings.

Well, you have offered two speculations and one error of fact. And you
accuse me of making a 'jerk of myself' for correctly answering his
question as written? Nitwit.

Great. The last paragraph is what I was looking for.

Thank you, DoN

On 11 Jul 2005 21:20:07 -0400, (DoN. Nichols)
wrote:

In article ,
McLeod wrote:
On 11 Jul 2005 08:30:09 -0700, "
wrote:

wrote:

Does anyone understand what are the mechanics behind the ISO settings
for RAW images?

Yes, I understand them. At least for the Canon 1D Mark II: there are
three buttons, a finger wheel, and a thumb wheel. Maybe the Nikon D70s
has the Telepathic Adaptor which can lock onto the mind of the
photographer, thus dispensing with these crude mechanical implements?

It depends on the camera. The Nikon D70 has an "AUTO-ISO" mode
(selected from the menu), in which, first the camera attempts to select
an appropriate shutter speed and aperture at the base ISO of 200, and if
that does not achieve the desired effect, it then increments the ISO
until a "reasonable" combination of shutter speed and aperture are
possible -- or until it gives up at the top ISO.

Menu entry 21 allows you to select a "floor" shutter speed, from
1/60 of a second down to 30 seconds, depending on what your needs are.
Yes, a higher floor would be nice with longer lenses, but with longer
focal lengths you are somewhat more likely to be using a Nikon VR lens
("Vibration Reduction" the equivalent of Cannon's IS "Image
Stabilization").

I think the poster was asking a more generalized question than you
responded to, but hey, who am I to stop you from making a jerk of
yourself.

Usually the lowest ISO setting of your digital camera is going to be
where you get the best picture with the lowest noise. Higher settings
are amplified, I believe, so noise is increased.

They are only in increments of 100, 200, 400 to remain reciprocal with
shutter speed and aperture settings.

Here, again, this is camera dependent. It may be so with your
Cannons, though I don't know, as I've not used one. However, the Nikon
D70 allows ISOs in the following steps:

200, 250, 320
400, 500, 640
800, 1000, 1250
1600

Rather finer steps than those which you listed, but the D70
adjusts exposures in 1/3 stop increments.

As for the mechanism of the actual ISO changes, it is
accomplished by one or more variable gain stages between the analog
sensor cells and the A/D converter(s) which feed the digital signals
into the camera's buffer memory. I don't know the count of A/D
converters, but I would expect multiple ones, each allocated to a
subgroup of sensors, so the A/D conversion could be accoplished at least
partially in parallel, to speed up the process, and thus to clear the
camera's sensor for the next shot more quickly.

Enjoy,
DoN.

On Tue, 12 Jul 2005 01:29:30 -0000, Jeremy Nixon
wrote:

wrote:

But how does it work for RAW in digital storage? "Approximation"
algorithms are not supposed to be applied and whatever light reaches
it should be recorded no matter what ISO is.

Any ideas?

The signal is amplified to reach the desired level.

Got it now.

Thank you.

In article ,
wrote:
Sorry, I realize now that the original message was not "descriptive"
enough.

How ISO works on film is rather "material" - the grain of ISO 400
compared to let's say 50, has a different, more corse structure.

But how does it work for RAW in digital storage? "Approximation"
algorithms are not supposed to be applied and whatever light reaches
it should be recorded no matter what ISO is.

Any ideas?

As I just got through typing (and thus, you have not yet had
time to read it), and perhaps I can add a bit more detail this time, now
that I know the focus of your question.

1) The sensor cells (CCD or CMOS) have capacitors which store a
charge during the exposure. The maximum change which can be
stored defines the low-end ISO for the sensor, as more exposure
fills the cell to the maximum, and highlight detail is lost.

2) Once the exposure is done, the charge is moved from the sensor
cells towards circuitry on the border of the sensor.

3) The first bit of circuitry will be a gain stage, so if the
exposure was not sufficient to fill *any* of the cells, the
range of voltage resulting from the charges which *were* stored
is increased by applying amplification, to provide a reasonable
range of voltage to the next stage.

4) The second bit of circuitry will be the A/D converter (Analog to
Digital converter.) This converts the analog information from
the sensor cell which came through the amplifier into a number
which represents that level in a digital form

Note that there is always *some* noise in the sensors, though it
may be small enough so it is not noticed in the presence of the signal
at low gains.

However, if the gain is high, it is amplifying both the noise
(which is approximately a constant level for a given exposure time) and
the signal. This brings the noise up to a level where it is obvious
(typically on darker areas of the photograph).

There are at least three sources of noise:

1) "Thermal" noise, which is a function of the resistance
(impedance) that the sensor and circuitry up to the amplifier
sections and the temperature. The higher the temperature, the
more thermal noise, and the higher the gain (ISO) the more
visible the noise from a given temperature will be. Far
Infrared sensors are often operated at very cold temperatures,
such as 70K (the temperature of liquid nitrogen). You've
probably seen far IR images in the video from missile strikes
from aircraft from the Gulf War footage.

2) The digital "clocking" noise (which is left-overs from the signals
used to move the information from the sensors to the A/D
converters.

3) "Fixed Pattern" noise, resulting from minute variations in the
sizes of the sensors, and slight variations in the accuracy of
the A/D converters.

This last one has the advantage of being somewhat predictable,
and many cameras have a mechanism to subtract this out from the
image data. This is commonly done (for long exposures) by
closing the shutter, and collecting noise for the same time as
the exposure just collected, and then digitally subtracting the
noise from the signal data.

The noise from these various sources all tends to be worse for
long exposures or high ISO, and as a result, it behaves in a manner
somewhat similar to film pushed to high ISOs. However, the film won't
have the Fixed Pattern or the clocking noise sources, just the
equivalent of the thermal noise. (There is a difference. As you push
the ISO in a digital sensor, the brightness variation becomes greater,
but the size remains that of a single pixel. As you push the ISO in
film, however, individual grains become much larger than the pixel size
from the digital image at the same ISO, so the noise (grain) from
seriously pushed film is more obvious than that from the sensors at
higher ISOs. Yes, you can see the noise when you push a Nikon D70 to
1600 ISO, but it is not nearly as obnoxious as the grain in film pushed
to the same speed.

Also, in digital images, there are various algorithms to reduce
the noise. Some are built into the cameras (including that for
correcting fixed-pattern noise), while other may be in the camera, or
may be in the computer during post-processing.

The D70 is not particularly aggressive in in-camera noise
reduction processing, and (from what I have read) the Cannon is much
more aggressive. One side effect is that the Cannon (at least the
Digital Rebel cameras) appears to produce softer images, as sharpness is
one of the casualties of aggressive noise reduction. I am rather glad
that the Nikon leaves it to post-processing (in the computer), so I can
make my own choices between sharpness and noise.

I'm not sure how much of this noise reduction may be done in a
Cannon camera in RAW mode.

I do know that this last 4th of July, I opted to shoot with the
in-camera long exposure noise reduction turned off, so I was not stuck
waiting for the same time as the last exposure to gather the noise
reduction information before I could shoot again. For most images, it
came out rather nice, but for some (when I held the shutter open for
longer than I expected, thanks to delays in the launching of the next
fireworks), I wound up with serious fixed-pattern noise in those images.

To see some selected examples, visit the following URL:

http://www2.d-and-d.com/EXAMPLES/july-4th/index.html

I hope that this helps,
DoN.
--
Email: | Voice (all times): (703) 938-4564
(too) near Washington D.C. | http://www.d-and-d.com/dnichols/DoN.html
--- Black Holes are where God is dividing by zero ---

DoN. Nichols wrote:
[]
There are at least three sources of noise:

1) "Thermal" noise, which is a function of the resistance
(impedance) that the sensor and circuitry up to the amplifier
sections and the temperature. The higher the temperature, the
more thermal noise, and the higher the gain (ISO) the more
visible the noise from a given temperature will be. Far
Infrared sensors are often operated at very cold temperatures,
such as 70K (the temperature of liquid nitrogen). You've
probably seen far IR images in the video from missile strikes
from aircraft from the Gulf War footage.

2) The digital "clocking" noise (which is left-overs from the signals
used to move the information from the sensors to the A/D
converters.

3) "Fixed Pattern" noise, resulting from minute variations in the
sizes of the sensors, and slight variations in the accuracy of
the A/D converters.
[]
I hope that this helps,
DoN

Don, thanks for this, but you have missed one important noise contribution
to the final image, which is photon-limited noise. In a nutshell, this is
a characteristic of light which produces an uncertainly in the number of
photons according to the square root of the number of photons collected.
So if the sensor well can only hold 10,000 electrons, there will be an
uncertainty of 100 electrons, and hence a best possible "signal-to-noise
ratio" of 100:1 at each pixel. Hence the desire for a large well size to
maximise SNR, and hence larger area sensors, and lower ISO settings.

Cheers,
David

In article ,
David J Taylor wrote:
DoN. Nichols wrote:
[]
There are at least three sources of noise:

1) "Thermal" noise, which is a function of the resistance

[ ... ]

2) The digital "clocking" noise (which is left-overs from the signals

[ ... ]

3) "Fixed Pattern" noise, resulting from minute variations in the

[ ... ]

Don, thanks for this, but you have missed one important noise contribution
to the final image, which is photon-limited noise. In a nutshell, this is
a characteristic of light which produces an uncertainly in the number of
photons according to the square root of the number of photons collected.
So if the sensor well can only hold 10,000 electrons, there will be an
uncertainty of 100 electrons, and hence a best possible "signal-to-noise
ratio" of 100:1 at each pixel. Hence the desire for a large well size to
maximise SNR, and hence larger area sensors, and lower ISO settings.

You have a good point -- though I'm not sure that I would class
this with "noise". It is at least an "uncertainty". And the fewer the
electrons, the more effect the uncertainty has.

Enjoy,
DoN.
--
Email: | Voice (all times): (703) 938-4564
(too) near Washington D.C. | http://www.d-and-d.com/dnichols/DoN.html
--- Black Holes are where God is dividing by zero ---

On 12 Jul 2005 20:11:12 -0400, (DoN. Nichols)
wrote:

In article ,
David J Taylor wrote:
DoN. Nichols wrote:
[]
There are at least three sources of noise:

1) "Thermal" noise, which is a function of the resistance

[ ... ]

2) The digital "clocking" noise (which is left-overs from the signals

[ ... ]

3) "Fixed Pattern" noise, resulting from minute variations in the

[ ... ]

Don, thanks for this, but you have missed one important noise contribution
to the final image, which is photon-limited noise. In a nutshell, this is
a characteristic of light which produces an uncertainly in the number of
photons according to the square root of the number of photons collected.
So if the sensor well can only hold 10,000 electrons, there will be an
uncertainty of 100 electrons, and hence a best possible "signal-to-noise
ratio" of 100:1 at each pixel. Hence the desire for a large well size to
maximise SNR, and hence larger area sensors, and lower ISO settings.

You have a good point -- though I'm not sure that I would class
this with "noise". It is at least an "uncertainty". And the fewer the
electrons, the more effect the uncertainty has.

Enjoy,
DoN.



The information you gave is far beyond I would've been able to gather
myself. The level of your understanding is amazing and you have my
greatest appreciation for the willingless to share.

Thank you all.

Regards,
Len

In article ,
wrote:
On 12 Jul 2005 20:11:12 -0400, (DoN. Nichols)
wrote:

[ ... all snipped at this point ... ]

The information you gave is far beyond I would've been able to gather
myself. The level of your understanding is amazing and you have my
greatest appreciation for the willingless to share.

Well ... it helps that I used to work with IR the development
and testing of IR imaging sensors, so I had plenty of time to absorb the
information. It was just a matter of re-describing them in terms of
visible light cameras.

Thank you all.

You're welcome,
DoN.
--
Email: | Voice (all times): (703) 938-4564
(too) near Washington D.C. | http://www.d-and-d.com/dnichols/DoN.html
--- Black Holes are where God is dividing by zero ---

DoN. Nichols wrote:
In article ,
David J Taylor
[]
Don, thanks for this, but you have missed one important noise
contribution to the final image, which is photon-limited noise. In
a nutshell, this is a characteristic of light which produces an
uncertainly in the number of photons according to the square root of
the number of photons collected. So if the sensor well can only hold
10,000 electrons, there will be an uncertainty of 100 electrons, and
hence a best possible "signal-to-noise ratio" of 100:1 at each
pixel. Hence the desire for a large well size to maximise SNR, and
hence larger area sensors, and lower ISO settings.

You have a good point -- though I'm not sure that I would class
this with "noise". It is at least an "uncertainty". And the fewer
the
electrons, the more effect the uncertainty has.

DoN, it's noise in the sense that if you photograph a uniform field, the
value of each pixel within the field will not be a single value, but will
have the photon-noise uncertainty superimposed on top of the nominal
value. Thus is looks just like the other noise sources you describe.

I can't remember now if the statistics of the noise are Gaussian (Poisson
rings a bell), and whether the square root is the RMS value of the noise.
I also used to work with IR sensors, by the way, cooled to 77K as I
recall. I modelled the entire vision system - target, atmosphere, sensor,
monitor, eyeball, brain.

Cheers,
David