Samsung sensor patent application

Yes, that's fair enough comment. I recall that when I had an SLR my
maximum was ISO 200 (film-limited), and with my compact cameras ISO 400
(when pushing things too far!), but if a penalty-free ISO 50,000 were
available, of course it would be included in my settings as well! I have
found the extra ISO range of the DSLR very useful.

Cheers,
David

 

That's describing shot noise.
That's what causes noise in shadows - not highlights.

 

Frederick, photon-limited noise causes more noise in the highlights than
the shadows, as there are more photons. The noise is proportional to the
square root of the number of photons, and there are more photons in the
highlights than in the shadows.

We appear to be at cross-purposes here. What do you think causes noise in
the highlights?

Cheers,
David

 

Ooops - rereading "noise due to readout inaccuracy in the
highlights may not matter as much" - yes I agree.

But that doesn't tie in with:
"Does the noise in the highlights matter that much? After all, it's
where the photon noise is the greatest"

 

frederick wrote:
[]

The confusion comes, perhaps, because although the numerical value of the
(shot) noise is greatest in the highlights (and hence shot noise is more
important than ADC noise or other inaccuracies, at least in a
well--designed system), the signal is greatest as well, and hence the
signal-to-noise ratio is highest.

Yes, it is in the shadows where the noise is most visible, but the noise
there is caused both by shot noise /and/ by read-out noise etc.

Any better?

Cheers,
David

 

When using my old SLRs, my upper limit was *usually* ISO 400. The
reason for the "usually" is that several days ago while getting rid
of old junk, I came across several unopened boxes containing 24 exp.
cassettes of Kodak's Ektar 1000. The expiration dates were 08/1990.
I don't recall ever shooting any of that Ektar, so ISO 400 may
really have been my upper limit!

 

"Photon-limited noise" is noise limited by photons. I
doubt you will find anything discussing that.

What you want to discuss is "photon noise limited",
which is a totally different beast.

But of course there is also more signal, proportionally,
and therefore the SNR is better and the noise has less
effect in the highlights.

There may be less photon noise in the shadows, but it is
much closer in level to the signal, and therefore is of
significance.

People shouting?

 

Yes. Photon noise limited merely means that the signal
is now large enough that photon noise is greater than
read noise. It does *not* mean that photon noise is
therefore a problem, it means that read noise is now
less of a problem.

 

Google provides 314 entries for "photon-limited noise", but I agree that
"photon-noise limited" is a better description, and provides over 15,000
hits.

... as I already said.

You have an insight here which I am, fortunately, lacking.

David

 

It's about the implementation. What Samsung appear to be proposing is
an "underlayer" of logic that sees the well hit "full", clears it, and
counts the event. You need this logic and counter for every pixel. It
is not a small undertaking...

As to words, it was how people interpreted your words then in the
context of the discussion then. Then "empty" meant bucket-brigading the
sensor to read it and empty the pixels.

Cheers,
Alan.

 

Alan Browne wrote:
[]

We share a common, but not identical language. It has caught out many
here.

Cheers,
David

 

But fortunately, you have a superior humor to noise ratio.

 

No sale.

 

Shot noise is greatest, in an absolute sense, in the highlights.
Relative to signal, it is greatest in the shadows. However, below a
certain exposure level, there is more read noise than shot noise, so shot
noise tends to be an issue that affects mainly higher and middle tones,
as the damage shot noise does to the deeper tones is nothing compared to
what read noise does down there.

 

In the context of which we are speaking, the noise threat will most
likely be contour bands of changing noise patterns, making it more
visible than noise in a simple SNR curve.

Again, what if the problem is accentuated along an exposure contour?

No doubt it will capture highlight details normally clipped; I just
wonder how many extra artifacts must be suffered for them, though.

I don't want them either, if they exist for the usual reason - low
quantum efficincy. If, however, you count wells over and over, or are
able to increase the capacity of photons per unit of area without
dropping QE, then you have basically also introduced lower ISOs. In the
Samsung patent model, you should be able to set the camera to ISOs lower
than the typical current camera if the wells can be recounted again and a
again; the difference between ISO 100 with 8 stops of headroom and ISO
3.125 with 3 stops of headroom is just a metering/logistics difference.

 

(Floyd L. Davidson) wrote in

Shot noise gets less and less relevant, compared to read noise, the
further you go into the shadows. Here's a model of a 5D pixel at ISO
100:

http://www.pbase.com/jps_photo/image/67443532/original

Read noise results in about 11% the SNR at 12 stops below saturation as
without any read noise (pure shot noise), and 1:1 SNR occurs about 4.5
stops higher with the read noise.

The notion that "current cameras are shot noise limited" needs a bit of
qualification. They are limited by shot noise only in the brighter
tones. They are limited in the shadows by read noise.

There are never enough photons for a clean image in extremely deep
shadows, or at extreme high ISOs, but the usability of such would be far
better without read noise.

Here's a simulation of pure shot noise compared to shot noise plus read
noise; img is the original noiseless image; cimgph is with poisson
statistics applied for a maximum of 19.53 photons at RAW saturation (our
whitepoint) at ISO 204,800; cimgphrd is with read noise added. The model
pixel here is one with 80,000 electrons at ISO 100 RAW saturation (3.5
stops above metered ISO 100's gray), with Canon 5D read noise of 26
electrons at ISO 100. This would correspond to a 5D with better
microlenses (the ones in the 5D are poor). This is a "100% crop"
(original pixels, grayscale or one CFA color channel):

http://www.pbase.com/jps_photo/image/94289478/original

Then again at ISO 3,279,000:

http://www.pbase.com/jps_photo/image/94289481

And again at ISO 52,630,00:

http://www.pbase.com/jps_photo/image/94289483

What limits the shadows?

 

Agreed - implementation dependant. Let's see what the actual results are.

I'm thinking of specular highlights rather the bride's dress, but agreed.

Yes, I agree. But given the choice of ISO 50 at super-duper quality and
ISO 200 with excellent quality, in most of my photography the ISO 200
would win every time, as I prefer compact and lightwieght and therefore
the f/4-5.6 zooms rather than the f/2.8 variety.

Cheers,
David

 

whenever I look at images,
the highlights are virtually noise free, all the time.

the shadows are the only place that noise is significant.

 

.... and ... ?

You left something out of your reply.

What you're saying is true, of course, at low ISOs and with big pixels.

My examples there are extremely high ISOs, from extremely pushed shadows
of "low ISOs". Ideally, a camera that only counted photons collected and
added no noise of its own would be somewhat usable at extremely high ISOs
or have much more usable shadows at low ISOs than current cameras with
extra camera-generated read noise (and thermal noise, too).

The point is, photon shot noise is not the limiting factor to extreme
shadows or extreme ISOs that it is often purported to be; it is read
noise that makes them almost totally useless.

 

Bob wrote:
[]

Yes, the signal-to-noise ratio is higher when the photon flux is greater.
However, the numerical value of the noise is higher.

Cheers,
David