How does ISO setting work?

"John Sheehy" wrote in message
...
"MikeWhy" wrote in
:

"John Sheehy" wrote in message
...
Paul Furman wrote in
- september.org:

Alpine sunset at dusk with beautiful bright sky and clouds:
(and almost all scenes with a sky or important highlights)
-keep ISO low and use a tripod when needed.

As computers/storage/firmware get faster, we could be shooting 16 ISO
1600 shots in rapid succession for an aligned ISO 100 composite, for
a less read
noise, and less camera shake blur, than "normal" ISO 100.

That's no bargain. That exposure is 4x as long as a single ISO 100
shot,

No; total exposure time is exactly the same. ISO is linear.

ISO 1600 is 4 stops down from ISO 100, not 16 stops.

not counting 15 additional starts and stops.

Well, this was a context where everything got faster ...
I believe that there is a new P& or 4/3 camera that does something like
this now. If it is truly a "tripod" type shot, then the extra time is not
much of an issue. You'll just have hyphenation of anything moving against
the background, though.

Please check the numbers; I'm pretty sure you have part of that wrong.
Stacking 8 exposures shot at (virtual) ISO 25600 might be what was intended.

Hyphenated, versus simply smeared in the single longer exposure. No biggee.


(The world is
divided into 10 kinds of people: those who get it, and those who
don't.)

Just a bit ...

100 * 2^4 = 1600.
100 * 2^8 = 25600.
100 * 2^16 = 6553600.

John Sheehy wrote:
Paul Furman wrote in -
september.org:

Closest Nikon has is to underexpose then lift the shadows.

Well, Nikon has a number of cameras with very "liftable" low-ISO shadows.
Canon is plagued with excessive shadow banding at low ISOs. The 7D has a
lot of banding at low ISOs, but much of it is fixed pattern (and some
apparently caused by poor firmware correction), and could have conceivable
be calibrated out. I can get a decent small web-image from my 7D at ISO
100 under-exposed 8 stops, after subtracting a stack of blackframes.

Be interested to see an example of shadow banding at low ISO on the 7D -
a fragment at full resolution PNG lossless compressed would do.

Regards,
Martin Brown

On 10-03-10 21:59 , Neil Harrington wrote:
snip I could be mistaken, but my impression is that he just wanted a
fairly simple answer to a fairly simple question.

Isn't it wonderful how we can take the simple and make it complex, but
we have a hard time taking the complex and making it simple?

--
gmail originated posts are filtered due to spam.

Martin Brown wrote in
:

Be interested to see an example of shadow banding at low ISO on the 7D
- a fragment at full resolution PNG lossless compressed would do.

I'll soon be making examples for a thread on DPReview; I'll try to link to
the images here.

I've actually seen the banding in OOC JPEGs in deep shadow areas, without
any pulling up of the shadows (ALO off).

There are basically two components to 7D low ISO banding; one is the
typical recent-Canon non-fixed banding with lots of low-frequency content.
The other is unique to the 7D, at least as a general problem. It is
potential deep modulation of every 8th column of pixels, and different
blackpoints for odd and even pixel rows. Some people are lucky, in that
none of the 8 vertical lines in the 8-line period are deep, but most 7Ds
have at least one period member that's pretty deep.

Canon does a very ****-poor calibration of blackpoint-by-line in their
cameras; Canon does not guard against the pitfalls of complex readout, and
hence, their Rebels have better shadow IQ than their semi-pro siblings with
similar sensors.

John Sheehy wrote:
Martin Brown wrote in
:

Be interested to see an example of shadow banding at low ISO on the 7D
- a fragment at full resolution PNG lossless compressed would do.

I'll soon be making examples for a thread on DPReview; I'll try to link to
the images here.

I've actually seen the banding in OOC JPEGs in deep shadow areas, without
any pulling up of the shadows (ALO off).

If you can find one of these OOC JPEGs I would be interested to take a
look at it. My curiosity is peeked. The most obvious cause would be some
kind of rounding error in the transfer function from RGB to YCC.

Please send one to me privately at the odd looking reply to address.
(no alterations needed it is valid when unmodified)

Certain other well known leading packages have very similar faults
although the damage is usually most annoying in highlights where clear
chroma to luminance bands can appear in highlights.

A green to magenta diagonal gradient put through the JPEG encode/decode
cycle a few times is a very good test of rounding errors in the codec.
any rounding errors and it will break up into chunky bars of colour.

There are basically two components to 7D low ISO banding; one is the
typical recent-Canon non-fixed banding with lots of low-frequency content.
The other is unique to the 7D, at least as a general problem. It is
potential deep modulation of every 8th column of pixels, and different
blackpoints for odd and even pixel rows. Some people are lucky, in that
none of the 8 vertical lines in the 8-line period are deep, but most 7Ds
have at least one period member that's pretty deep.

Unfortunate given that JPEG relies on 8x8 periodicity.

Canon does a very ****-poor calibration of blackpoint-by-line in their
cameras; Canon does not guard against the pitfalls of complex readout, and
hence, their Rebels have better shadow IQ than their semi-pro siblings with
similar sensors.

I only have Canon Ixus P&S but I can't say I have ever had problems with
their black point calibration. Its main attraction is always being with me.

Regards,
Martin Brown

In rec.photo.digital John Sheehy wrote:
Paul Furman wrote in -
september.org:

Closest Nikon has is to underexpose then lift the shadows.

Well, Nikon has a number of cameras with very "liftable" low-ISO shadows.
Canon is plagued with excessive shadow banding at low ISOs. The 7D has a
lot of banding at low ISOs, but much of it is fixed pattern (and some
apparently caused by poor firmware correction), and could have conceivable
be calibrated out. I can get a decent small web-image from my 7D at ISO
100 under-exposed 8 stops, after subtracting a stack of blackframes.

Doesn't Nikon black clamp its images? If so for a fair comparison the
Canon images should be manually black clamped to the same extent.

--
Chris Malcolm

Martin Brown wrote in
:

If you can find one of these OOC JPEGs I would be interested to take a
look at it. My curiosity is peeked. The most obvious cause would be
some kind of rounding error in the transfer function from RGB to YCC.

I didn't say anything about the problem *being* a JPEG issue. This is
purely a RAW thing. What I did say is that it could be seen in the JPEG
review image, meaning, with no pushing of the shadows. This is purely a
RAW shadows issue.

Chris Malcolm wrote in
:

Doesn't Nikon black clamp its images? If so for a fair comparison the
Canon images should be manually black clamped to the same extent.

Black-clipping the RAW data does not remove noise; that is a popular myth.
It removes more signal than it does noise, lowering SNR ever so slightly.
Most converters seem to clip Canon RAW data as soon as it loads, anyway.

In fact, offsetted RAW data loses more noise with downsampling. When I do
manual conversions step by step, I always leave Canmon RAW data with
negative values near black, after subtracting the offset, and don't black-
clip until it's absolutely necessary (display image). The negative values
balance out the positive noise in any kind of noise filtering or
downsampling, bringing means near black closer to linearity. In black-
clipped RAW data, you are left with blacks that have means above black,
different for the three channels after WB; a mess, which many companies
deal with by just trashing the deep shadows by removing color, and/or
blackening near-blacks with a tone curve.

In rec.photo.digital John Sheehy wrote:
Chris Malcolm wrote in
:

Doesn't Nikon black clamp its images? If so for a fair comparison the
Canon images should be manually black clamped to the same extent.

Black-clipping the RAW data does not remove noise; that is a popular myth.
It removes more signal than it does noise, lowering SNR ever so slightly.
Most converters seem to clip Canon RAW data as soon as it loads, anyway.

In fact, offsetted RAW data loses more noise with downsampling. When I do
manual conversions step by step, I always leave Canmon RAW data with
negative values near black, after subtracting the offset, and don't black-
clip until it's absolutely necessary (display image). The negative values
balance out the positive noise in any kind of noise filtering or
downsampling, bringing means near black closer to linearity. In black-
clipped RAW data, you are left with blacks that have means above black,
different for the three channels after WB; a mess, which many companies
deal with by just trashing the deep shadows by removing color, and/or
blackening near-blacks with a tone curve.

I said "clamping" rather than "clipping" but I may have used the wrong
term. What I meant was I thought Nikon had decided on a black
threshold below which they would simply make the pixels completely
black. So the deepest shadows are perfectly black with no noise at
all.

--
Chris Malcolm

In article , MikeWhy
writes
"John Sheehy" wrote in message
...
"MikeWhy" wrote in
:

"John Sheehy" wrote in message
...

As computers/storage/firmware get faster, we could be shooting 16 ISO
1600 shots in rapid succession for an aligned ISO 100 composite, for
a less read
noise, and less camera shake blur, than "normal" ISO 100.

That's no bargain. That exposure is 4x as long as a single ISO 100
shot,

No; total exposure time is exactly the same. ISO is linear.

ISO 1600 is 4 stops down from ISO 100, not 16 stops.

And 4 stops is 1/16th of the exposure, so 16 shots at ISO1600 is the
same exposure time as 1 shot at ISO100.

You seem to be confusing stops with # of exposures.

100 * 2^4 = 1600.

Correct and put another way this shows:
2^1 (ie. 2) shots at ISO1600 is the same exposure time as one ISO800
2^2 (ie. 4) shots at ISO1600 is the same exposure time as one ISO400
2^3 (ie. 8) shots at ISO1600 is the same exposure time as one ISO200
2^4 (ie. 16) shots at ISO1600 is the same exposure time as one ISO100

100 * 2^8 = 25600.
So 256 exposures at ISO25600 will take the same time as one at ISO100.

100 * 2^16 = 6553600.
So 65536 exposures at ISO6553600 will take the same time as one at
ISO100.
--
Kennedy
Yes, Socrates himself is particularly missed;
A lovely little thinker, but a bugger when he's ****ed.
Python Philosophers (replace 'nospam' with 'kennedym' when replying)