Another ISO question...

John Sheehy wrote:
"Roger N. Clark (change username to rnclark)" wrote
in :

For example,
in a Canon 30D, the maximum signal at ISO 100 corresponds to
about 50,000 photons (which are converted into 50,000 electrons).
Noise = square root 50000 = 223.6 electrons,
signal / noise = 50000/223.6 = 223.6.

Actually, there isn't enough headroom for 50000 photons at ISO 100.
There are about 25000 at ISO 200 RAW saturation, but only about 44000
total photons possible at ISO 100 on the 20D, which has the same sensor.

Not on the cameras I've tested. Comes out 51,000. The 10D was
44,000.

At iso 1600, the 30D maximum signal is amplified to give:
50000/16 = 3125 photons (approximately). The noise is sqrt(3125)
=55.9, and signal / noise = 55.9.

Because digital cameras have noise dominated by photon noise (that's
the best one can do), a number of properties can be predicted
about digital camera performance.

For example:
http://www.clarkvision.com/imagedeta...el.size.matter

http://www.clarkvision.com/imagedeta...formance.summa
ry

The f/ratio Myth and Digital Cameras
http://www.clarkvision.com/photoinfo/f-ratio_myth

The Depth-of-Field Myth and Digital Cameras
http://www.clarkvision.com/photoinfo/dof_myth
I am finding that in lowlight I can get a pretty decent exposure in
terms of digital noise at around 400 ISO, sometimes even 800
Yes, this is well predicted. See the "Does Pixel Size Matter"
article above. Not all cameras will have good high ISO performance,
and Canon is not the only manufacturer to have good high ISO
performance. (Hint: it has more to do with pixel size rather than
camera manufacturer or CMOS versus CCD.)

Image noise only has to do with pixel size when you're comparing cameras
with the same MP count. A large sensor with more and smaller pixels
doesn't necessarily have any more image noise than a large sensor with
large pixels, and has more resolution. Pixels from the 1.97 micron FZ50,
binned, outperform every DSLR Pixel at low ISO, and only underperform the
cleanest Canons at ISO 1600 by a small amount, and outperform all DSLRs
from Nikon, Olympus, Pentax, etc. Image noise (not pixel noise) remains
mainly unchanged with binning, if the image is still viewed at the same
size (IOW, binning is not really necessary for reduced image noise).

Well, this is just plain stupid. Look at your logic: you have
a 1.97 micron pixel pitch on the FZ250, 10 megapixels better than all
DSLRs from Nikon etc, and all the canons except at ISO 1600?
Unity gain ISO on the FZ50 must be below 100. Compare the results
from a Canon 5D to an FZ50. No contest. For what you say to
be even close, theoretically you would need 17 FZ250 pixels for every
single Canon 5D pixel. The FZ50 would then need to be a 170 megapixel
camera!

Roger

David J Taylor wrote:
the_niner_nation wrote:
[]
and also, just how *good* are the post processing programs people use
at reducing ' digital noise' ?
I have heard that having an IS feature on a lens can make as much as
3 fstops worth of difference in helping to attain a decent
shake-free image in low light/smaller aperture settings..is it as
high as 3 fstops or is that a bold claim?

On some of the Panasonic Lumix cameras, yes. On some DSLR lenses perhaps
a little less (the glass is heavier and not so easy to move dynamically).

Only a small internal element is used to shift the image.
You don't need to move a heavy element.
Here is an example of a 500 mm f/4 lens with a 2x teleconverter
at 1/10 second:
http://www.clarkvision.com/galleries...1500b-700.html

Here is another case of hand-holding a large 500 mm f/4 IS lens:
http://www.clarkvision.com/galleries...598.b-700.html
While the exposure time is high (1/1600) second, a colleague
was shooting with a 300 mm f/2.8 not stabilized and got no sharp
images, even with faster exposure times.

Roger

"Roger N. Clark (change username to rnclark)" wrote
in :

Not on the cameras I've tested. Comes out 51,000. The 10D was
44,000.

You might want to check the linearity of the highlights on both of those
cameras.

My 20D goea all the way up to 4095 in the RAW data at ISO 100, but if you
compare to an ISO 200 image, the RAW levels are the same throughout the
shadows and highlights, but at the high end the RAW values go
increasingly too high at ISO 100, and clips at a lower point in a
highlight gradient. My 10D is just the opposite. It rolls in extra
highlights just below RAW saturation at ISO 100. You can test these
tings by shooting an OOF gradient of a light falling off along a wall,
and make split images at ISO 100 and 200 with half the exposure time.
After you calibrate for any small differences in global RAW levels, you
might see that the images differ only in extreme highlights; the 10D
capturing more highlights before clipping at ISO 100, and the 20D
capturing less at ISO 100.

--


John P Sheehy

"Roger N. Clark (change username to rnclark)" wrote
in :

John Sheehy wrote:

Image noise only has to do with pixel size when you're comparing
cameras with the same MP count. A large sensor with more and smaller
pixels doesn't necessarily have any more image noise than a large
sensor with large pixels, and has more resolution. Pixels from the
1.97 micron FZ50, binned, outperform every DSLR Pixel at low ISO, and
only underperform the cleanest Canons at ISO 1600 by a small amount,
and outperform all DSLRs from Nikon, Olympus, Pentax, etc. Image
noise (not pixel noise) remains mainly unchanged with binning, if the
image is still viewed at the same size (IOW, binning is not really
necessary for reduced image noise).

Well, this is just plain stupid. Look at your logic: you have
a 1.97 micron pixel pitch on the FZ250, 10 megapixels better than all
DSLRs from Nikon etc, and all the canons except at ISO 1600?

I'm not sure I quite understand your language there. What you are
paraphrasing is not what I wrote, at all.

Unity gain ISO on the FZ50 must be below 100.

Unity gain is a meaningless concept, in the face of analog noise. There
is no meaningful relationship between the charge of an electron and an
ADU. An ADU has a totally arbitrary value.

Compare the results
from a Canon 5D to an FZ50. No contest.

If you compare a crop from the 5D sensor the size of an FZ50 sensor, then
there is no contest. The only area the 5D is going to excel in is image
read noise at ISOs 800 and 1600. The FZ50 crop will have less read noise
at ISO 100, and slightly less shot noise, image-wise (bin/downsample or
look at the fine noise with all the extra detail - your choice).

For what you say to
be even close, theoretically you would need 17 FZ250 pixels for every
single Canon 5D pixel.

16.5, I figure.

The FZ50 would then need to be a 170 megapixel
camera!

Yes; wouldn't that be lovely?



--


John P Sheehy

John Sheehy wrote:
"Roger N. Clark (change username to rnclark)" wrote
in :

John Sheehy wrote:

Image noise only has to do with pixel size when you're comparing
cameras with the same MP count. A large sensor with more and smaller
pixels doesn't necessarily have any more image noise than a large
sensor with large pixels, and has more resolution. Pixels from the
1.97 micron FZ50, binned, outperform every DSLR Pixel at low ISO, and
only underperform the cleanest Canons at ISO 1600 by a small amount,
and outperform all DSLRs from Nikon, Olympus, Pentax, etc. Image
noise (not pixel noise) remains mainly unchanged with binning, if the
image is still viewed at the same size (IOW, binning is not really
necessary for reduced image noise).

Well, this is just plain stupid. Look at your logic: you have
a 1.97 micron pixel pitch on the FZ250, 10 megapixels better than all
DSLRs from Nikon etc, and all the canons except at ISO 1600?

I'm not sure I quite understand your language there. What you are
paraphrasing is not what I wrote, at all.

Unity gain ISO on the FZ50 must be below 100.

Unity gain is a meaningless concept, in the face of analog noise. There
is no meaningful relationship between the charge of an electron and an
ADU. An ADU has a totally arbitrary value.

Compare the results
from a Canon 5D to an FZ50. No contest.

If you compare a crop from the 5D sensor the size of an FZ50 sensor, then
there is no contest. The only area the 5D is going to excel in is image
read noise at ISOs 800 and 1600. The FZ50 crop will have less read noise
at ISO 100, and slightly less shot noise, image-wise (bin/downsample or
look at the fine noise with all the extra detail - your choice).


I don't understand this part, I assume we are talking about binning
enough pixels on the FZ50 so that the area comes out to the same on the
5D? If so then if the read noise on the FZ50 was the same on a pixels
basis as the 5D binning 16 pixels would increase the read noise by a
factor of 4. Figure it this way, you capture say 4000 electrons over
your 16 pixels, the 5D capture 4000 electrons on its one pixel. Say both
have a read noise of 5 electrons, the 5D will have a read noise of 5 out
of 4000, but the binned pixels from the FZ50 will have a read noise of
20 out of 4000.

But the 16 pixels from the FZ50 are not likely to even capture as many
electrons as the 5D, since the fill factor is likely to be a lot less.

Scott

Roger N. Clark (change username to rnclark) wrote:
[]
Only a small internal element is used to shift the image.
You don't need to move a heavy element.
Here is an example of a 500 mm f/4 lens with a 2x teleconverter
at 1/10 second:
http://www.clarkvision.com/galleries...1500b-700.html

Here is another case of hand-holding a large 500 mm f/4 IS lens:
http://www.clarkvision.com/galleries...598.b-700.html
While the exposure time is high (1/1600) second, a colleague
was shooting with a 300 mm f/2.8 not stabilized and got no sharp
images, even with faster exposure times.

Roger

Agreed, but it's likely that the moving element will be smaller on the
smaller lens, than on the bigger lens. The focal length ratio between
DSLR and compact camera can be around 5:1. Your examples are certainly
impressive (as I have come to expect!).

Cheers,
David

On Jul 11, 7:36 pm, "Roger N. Clark (change username to rnclark)"
wrote:
Ray Macey wrote:
On Jul 12, 6:21 am, "the_niner_nation"
wrote:
Ok, more noob 'ISO' questions...

how many psuedo-equivalent 'fstops' does increasing the ISO give when
trying to shoot a picture in poor lighting conditions?

Every doubling of the ISO is an extra stop. Similar process to
shutter speed in that regard

I am fully aware that increasing the ISO isn't *actually* raising the fstop
at all, it merly allows more light to hit the camera sensor, but talking

Actually, it doesn't change the amount of light hitting the sensor, it
merely raises the sensitivity of the sensor, so that it's better at
detecting the light that does hit it. This increased sensitivity is
the direct cause of the extra noise that is the downside of increasing
the ISO

Actually, it doesn't change the sensitivity of the sensor at all.
Changing ISO doesn't change the amount of light (the number of
photons) encountering the sensor, and doesn't change the number of
photons the sensor captures. This is true whether you operate at
IOS 100 or 1600. If the sensor captures 1000 photons at ISO 100
in a 1/100 second exposure, at ISO 1600 1/100 exposure, the sensor
captures the same 1000 photons.

What changes is the gain that is applied to the signal that
the photons generate. In the above example, the camera at
ISO 1600 exposure getting 1000 photons simply amplifies
the signal the photons generate by a factor of 16 more than
the camera would at ISO 100 for the same 1000 photons.

Further, the noise in the two signals in the above example
is essentially the same. The noise is the square root of the
number of photons collected plus some read noise (a few electrons).
a converted photon generates on electron, so in a 1000 photon signal,
the noise is dominated by the noise in the arrival rate of the
photons, not amplifier or other camera electronics.

So why you see more noise in the ISO 1600 example, versus ISO 100
is that the signal and the photon noise is amplified more in the
higher ISO. So really in the high ISO signal, you are looking at and
amplifying the lower signals (lower number of photons. For example,
in a Canon 30D, the maximum signal at ISO 100 corresponds to
about 50,000 photons (which are converted into 50,000 electrons).
Noise = square root 50000 = 223.6 electrons,
signal / noise = 50000/223.6 = 223.6.

At iso 1600, the 30D maximum signal is amplified to give:
50000/16 = 3125 photons (approximately). The noise is sqrt(3125) =55.9,
and signal / noise = 55.9.

Because digital cameras have noise dominated by photon noise (that's
the best one can do), a number of properties can be predicted
about digital camera performance.

For example:http://www.clarkvision.com/imagedeta...el.size.matter

http://www.clarkvision.com/imagedeta...erformance.sum...

The f/ratio Myth and Digital Camerashttp://www.clarkvision.com/photoinfo/f-ratio_myth

The Depth-of-Field Myth and Digital Camerashttp://www.clarkvision.com/photoinfo/dof_myth



I am finding that in lowlight I can get a pretty decent exposure in terms of
digital noise at around 400 ISO, sometimes even 800

Yes, this is well predicted. See the "Does Pixel Size Matter"
article above. Not all cameras will have good high ISO performance,
and Canon is not the only manufacturer to have good high ISO
performance. (Hint: it has more to do with pixel size rather than
camera manufacturer or CMOS versus CCD.)

Roger- Hide quoted text -

- Show quoted text -

Atta Boy Roger...

One for the "Clip and Save" file.

Jay Beckman
Chandler, AZ
www.pbase.com/flyingphotog

"the_niner_nation" wrote in message
.. .
Ok, more noob 'ISO' questions...

how many psuedo-equivalent 'fstops' does increasing the ISO give when
trying to shoot a picture in poor lighting conditions?

I am fully aware that increasing the ISO isn't *actually* raising the
fstop at all, it merly allows more light to hit the camera sensor, but
talking more generally in terms of allowing a faster shutter speed as a
trade off for the inevitible digital 'noise' on the final image...

and also, just how *good* are the post processing programs people use at
reducing ' digital noise' ?

I have heard that having an IS feature on a lens can make as much as 3
fstops worth of difference in helping to attain a decent shake-free image
in low light/smaller aperture settings..is it as high as 3 fstops or is
that a bold claim?

I am finding that in lowlight I can get a pretty decent exposure in terms
of digital noise at around 400 ISO, sometimes even 800


--
Posted via a free Usenet account from http://www.teranews.com


Just wanted to say thank you to everyone...this has easily been the most
useful,helpful and informative bunch of replies I have ever had...sincerly
appreciated!



--
Posted via a free Usenet account from http://www.teranews.com

Scott W wrote in
:

John Sheehy wrote:

If you compare a crop from the 5D sensor the size of an FZ50 sensor,
then there is no contest. The only area the 5D is going to excel in
is image read noise at ISOs 800 and 1600. The FZ50 crop will have
less read noise at ISO 100, and slightly less shot noise, image-wise
(bin/downsample or look at the fine noise with all the extra detail -
your choice).

I don't understand this part, I assume we are talking about binning
enough pixels on the FZ50 so that the area comes out to the same on
the 5D? If so then if the read noise on the FZ50 was the same on a
pixels basis as the 5D binning 16 pixels would increase the read noise
by a factor of 4.

Yes; in electrons, or absolute intensity. Relative to the new super-
pixel's photon capture, though, it is only 1/4 as strong.
Signal:ShotNoise has increased 4x, or 2 stops.

Figure it this way, you capture say 4000 electrons
over your 16 pixels, the 5D capture 4000 electrons on its one pixel.
Say both have a read noise of 5 electrons, the 5D will have a read
noise of 5 out of 4000, but the binned pixels from the FZ50 will have
a read noise of 20 out of 4000.

The fact of the matter is, however, that at ISO 100, the FZ50 original
pixels have about 3.25 electrons of read noise (slightly less at ISO
200), and the 5D has about 29 electrons of read noise, and 3.25 times 4
is still a bit less than 29. Over a stop less.

But the 16 pixels from the FZ50 are not likely to even capture as many
electrons as the 5D, since the fill factor is likely to be a lot less.

It isn't. The FZ50 in 16.5 pixels captures the same number of maximum
photons as the 5D; ~80,000. The FZ50 captures them faster, though, with
a higher quantum efficiency! At ISO 100, the FZ50 captures about 80K in
16.5 pixels, as compared to 52K in the 5D.

The horror stories about the inefficiencies of tiny pixels are greatly
overstated. The real "horror" comes from the size of the sensors they
are historically found in; the FZ50's sensor is a good size for easy lens
design, but it does not capture a lot of photons total, even though it is
fairly efficient per unit of area. It is unreasonable to declare that
tiny-sensor images are noisier than DSLR images because of small pixels;
the real reasons are the smaller total number of photons collected, and
lack of optimized high-ISO readout on P&S cameras. Even so, the FZ50 has
a higher max-signal-to-read-noise ratios than a Nikon D2X, at all ISOs.
The FZ50 has a higher SNR than the DX2, in the shadow areas, at the pixel
level, without any binning for the FZ50.

I am not advocating binning, per se. I'd rather have the original, full-
res image with all of its fine noise. I present the binned super-pixel
as a bridge concept, as it is something that one can grasp even if one
doesn't appreciate the concept that the power of noise in a displayed
image is not dependent only upon traditional noise statistics, but upon
the displayed size of the original pixel as well. They are two factors
affecting the subjective experience of noise. Another factor is the the
sharpness and contrast of the actual signal; the more easily it is
perceived, the less relevant the noise tends to be.

Most notions of binning giving better IQ come from contexts where the
result of the binning is viewed at the same pixel size as the original.
Even when they *are* displayed at the same subject size, if the original
resolution is viewed with any downsizing, the Nearest Neighbor algorithm,
or a hybrid thereof is often used which drops the influence of some
pixels, cancelling less noise. Nearest Neighbor maintains pixel-level
noise while reducing the number of pixels, a bad handicap for an image
downsized that way (NN actually increases it at the nyquist in the first
halving of resolution, with Bayer CFA cameras).

Take a crop of any noisy image in PS, duplicate a few copies of it, and
pixelate the copies 2x2, 3x3, 4x4, etc. The pixelated ones do not really
have less noise - the noise just gets coarser and has smoother gradients
to it. To me, that's worse - I'd rather see the fine noise and the
greater detail; my brain knows what it is, and knows to stop looking for
further details, and is happily focused. The quest for reduced pixel-
level noise has two main paths; a useful one where noise is reduced while
all else remains equal or better, and a useless one, entailing binning
and big pixels (which are actually a mechanical form of binning photons),
which is much like a dog chasing after its own tail, IMO.

--


John P Sheehy

On Thu, 12 Jul 2007 06:06:54 GMT, David J Taylor wrote:

I have heard that having an IS feature on a lens can make as much as
3 fstops worth of difference in helping to attain a decent
shake-free image in low light/smaller aperture settings..is it as
high as 3 fstops or is that a bold claim?

On some of the Panasonic Lumix cameras, yes. On some DSLR lenses perhaps
a little less (the glass is heavier and not so easy to move dynamically).

I don't think so. Nikon claims a 3 f/stop improvement with their
lightweight 55-200mm VR DX lens, and a 4 f/stop improvement with the
larger, heavier full frame 70-300mm VR II lens. From what I've seen
of the 55-200mm VR lens, Nikon sure isn't overstating the 3 f/stop
improvement.