The Joy of Pixel Density

Started Jul 13, 2008 | Discussions
aclo Regular Member • Posts: 480
Re: More shot noise

Tristan Cope wrote:

aclo wrote:

Tristan Cope wrote:

Yes. Lets take the extreme case and say that one photon is falling on
each sensor element. Signal = 1, noise = sqrt 1 = 1. Signal and noise
are equal (essentially photon flux per sensor element is entirely
random regardless of the signal). This is a quantum effect.

That the signal and noise have the same magnitude is irrelevant, as I
explained before (and others do later). And it also happens with
classical objects being randomly deposited onto a surface (such as
epitaxial growth), so it's not quantum (except inasmuch as the
photons being discrete is a "quantum" effect). Anyway this is off
topic.

My understanding is that shot noise is a quantum effect and caused by
the random nature of quantum level events.

It happens also with raindrops or indeed any kind of particles that are randomly deposited on some surface. On the other hand, that photons are discrete is indeed a "quantum" effect. Anyway, this isn't really important, it's just semantics.

OK think of this: imagine a photosite split in 10x10 smaller squares.
Suppose I "detect" photons separately in each, and there is no read
noise. Do you not see that shot noise will be unaffected by whether I
a) write out all 100 values, then add them, or b) I have a computer
(or analog counter) do it? (think about this a bit before dismissing
it). [and the photons aren't coherent so we really can argue like
above].

I entirely accept this, as long as the binning is done in hardware
before any signal amplification. Then it doesn't matter where the
photons fall within the binned area as long as they are counted.

I hope the idea gets across with my example.

Yes, thanks.

I still have a problem with the idea of binning in software though -
I don't see how this is really any different to downsizing a bitmap.

OK, if I have a signal of magnitude S (in one pixel; the units don't matter so long as they are the same for all quantities), then there are generically two different contributions to the noise: one that is constant (ie independent of the signal), call it R, and one that is proportional to the square root of the signal, call it N, the magnitude of which=sqrt(S).

Now the idea (from what I understand from reading a few links and posts, I in fact have no clue about electronics) is that if I bin nxn pixels in hardware then R will not change; of course, S will increase to n^2*S while the shot noise will become n*N=n*sqrt(S). So total noise sqrt(R^2+n^2 S) (uncorrelated noise sources add like this, under some assumptions which are valid here).

On the other hand, if I bin them after readout, then the signal is again n^2 S while the noise will be sqrt(n^2 R^2+n^2 S).

So you see, if I go to smaller pixels so that I must bin nxn of them to get the same size as the larger ones, then, if I do it in software, I'd need n times smaller read noise to get the same signal to noise. This ignores the CFA, and also aliasing issues.

By the way it's not different than downsizing a bitmap.

ejmartin Veteran Member • Posts: 6,274
Re: 400D Dynamic range not sensor limited

DSPographer wrote:

I didn't mean to put words in your mouth. I was just disappointed
that you went to the trouble to do this comparison that I find
uninteresting. If you have done the same comparison as this one but
at maximum ISO gain then I would very much like to see it. Can I
trouble you to link to it one more time? I do appreciate the effort
you make to do these comparisons (I don't have an FZ50 so I can't do
these tests myself) but I was frustrated that they didn't show the
sensor performance comparisons that I think are important. My post
above came out more disparaging then I intended, I just meant to
explain why I think maximum ISO noise performance is most important
for comparing sensors with different size pixels.

I think this may be what you're looking for:

http://www.pbase.com/jps_photo/image/74020772

moreover, it's the test that you and I think is more appropriate (at ISO 1600). The only drawback is that the 10D read noise is about 9 electrons, about a stop worse than the best current models which are around 4 electrons. So imagine the 10D image with half the noise...

-- hide signature --
OP John Sheehy Forum Pro • Posts: 21,729
Re: 400D Dynamic range not sensor limited

dethis wrote:

Nice demonstration John !!. Thanks.

In order to normalize for the 400D's 1stop more headroom i would like
the same demonstration but with ISO settings at
a) 100 for FZ50 vs 200 for 400D
b) 200 for FZ50 vs 400 for 400D

Well, I'd like to get it even more exact than that; I'd like to do cameras at base ISO and test DR by scaling the exposure so that each tonality is the same percentage of RAW saturation in all cameras involved. That takes a little more work, as I'll have to use some kind of neutral density material over the lighting to vary the quantity without varying the quality.

From what I remember of past measurements, the true base ISO of the 400D is about 87, and the true base ISO of the FZ50 is about 190 to 200. I didn't wind up with that ratio here, as I chose a whitepoint of 30 ADU for the FZ50, and 18 ADU for the 400D, to make them look equally bright, but the mean levels in the sky are not the same, hence the discrepancy. 14 or 15 ADU for the 400D whitepoint would have made the means look the same, but some of the "salt" of the noise makes things look brighter than they are, so the 400D crop would look brighter and noisier had the means been adjusted exactly.

Of course, I am never going to do the same scene as before; it will be a whole new scene. I just wish I had a D3 or a 1D3 to show just how close the FZ50 pixels can get to the best pixels out there, one a per-area basis, at ISO 1600, and how the FZ50 pixels are superior at low ISO. I have a feeling that most people would assume that the D3 and 1D3 would do better in these per-area comparisons, but actually, I expect them only to do better in noise (obviously not resolution) at ISO 1600 in the shadows. In the shot noise per unit of area, the FZ50 pixels should outperform all DSLRs at all ISOs.

-- hide signature --

John

OP John Sheehy Forum Pro • Posts: 21,729
Re: 400D Dynamic range not sensor limited

ejmartin wrote:

I think this may be what you're looking for:

http://www.pbase.com/jps_photo/image/74020772

moreover, it's the test that you and I think is more appropriate (at
ISO 1600). The only drawback is that the 10D read noise is about 9
electrons, about a stop worse than the best current models which are
around 4 electrons. So imagine the 10D image with half the noise...

And also imagine that I had chosen the whitepoints better and the 10D crops were brighter, in which case they would show more noise than they do. Also, I learned after I did this that the FZ50 has a botched PGA, which adds extra absolute noise, which makes 1600 worse than 100 at -4EC. Absolute read noise seems lowest at IS0 200 with the FZ50, and at 800 through 3200 on the 10D.

-- hide signature --

John

OP John Sheehy Forum Pro • Posts: 21,729
Re: 400D Dynamic range not sensor limited

John Sheehy wrote:

ejmartin wrote:

I think this may be what you're looking for:

http://www.pbase.com/jps_photo/image/74020772

moreover, it's the test that you and I think is more appropriate (at
ISO 1600). The only drawback is that the 10D read noise is about 9
electrons, about a stop worse than the best current models which are
around 4 electrons. So imagine the 10D image with half the noise...

And also imagine that I had chosen the whitepoints better and the 10D
crops were brighter, in which case they would show a tad more noise than
they do. Also, I learned after I did this that the FZ50 has a
botched PGA, which adds extra absolute noise, which makes 1600 worse
than 100 at -4EC. Absolute read noise seems lowest at IS0 200 with
the FZ50, and at 800 through 3200 on the 10D.

-- hide signature --

John

ejmartin Veteran Member • Posts: 6,274
Re: 400D Dynamic range not sensor limited

John Sheehy wrote:

In the shot noise per unit of area, the
FZ50 pixels should outperform all DSLRs at all ISOs.

John, you've stated that your FZ50 performs better in terms of noise for a given absolute exposure at ISO 100 pushed four stops than it does at ISO 1600, which is the reverse of any other camera I know of. It also flies in the face of the notion that the higher ISO amplification tends to reduce noises when referred back to input quantities like photoelectrons. I believe you called it a "Panasonic engineering blooper". Could it be reflective of a problem with the raw data being massaged by the camera at low ISO? If so that would seriously affect your conclusions; for instance, any sort of NR performed before writing the raw data would throw off the inferred sensor efficiency, by making it look more efficient than it is. As it stands, this anomaly in ISO 100 vs 1600 is troublesome, because as I said higher ISO amplification can only improve input-referred noise performance.

-- hide signature --
ejmartin Veteran Member • Posts: 6,274
Re: 400D Dynamic range not sensor limited

John Sheehy wrote:

I learned after I did this that the FZ50 has a
botched PGA, which adds extra absolute noise, which makes 1600 worse
than 100 at -4EC. Absolute read noise seems lowest at IS0 200 with
the FZ50, and at 800 through 3200 on the 10D.

Yes, and that seems to me to indicate that there is something fishy about the raw data coming from the camera; see my post below

http://forums.dpreview.com/forums/read.asp?forum=1018&message=28692270

Increasing the gain of a PGA only decreases contributions to noise of downstream sources when referred back to input values, because they get divided by the gain when referring them back; and upstream noise sources should be independent of ISO.

-- hide signature --
OP John Sheehy Forum Pro • Posts: 21,729
Re: 400D Dynamic range not sensor limited

ejmartin wrote:

John Sheehy wrote:

In the shot noise per unit of area, the
FZ50 pixels should outperform all DSLRs at all ISOs.

John, you've stated that your FZ50 performs better in terms of noise
for a given absolute exposure at ISO 100 pushed four stops than it
does at ISO 1600, which is the reverse of any other camera I know of.
It also flies in the face of the notion that the higher ISO
amplification tends to reduce noises when referred back to input
quantities like photoelectrons. I believe you called it a "Panasonic
engineering blooper". Could it be reflective of a problem with the
raw data being massaged by the camera at low ISO?

There are no signs of it; the sigma doubles with each 2x2 binning, the noise is nice and sharp as you can see.

If so that would
seriously affect your conclusions; for instance, any sort of NR
performed before writing the raw data would throw off the inferred
sensor efficiency, by making it look more efficient than it is. As
it stands, this anomaly in ISO 100 vs 1600 is troublesome, because as
I said higher ISO amplification can only improve input-referred noise
performance.

The proof is in the pudding, though, and that is why I used a real scene instead of a color checker - there is considerable detail 9 and 10 stops below saturation here; would that be possible with "massaged noise"? When you look at an ISO 1600 Sony A700 image, the shadows look all smoothed out. We have nothing like that here.

Oh, and BTW, the difference is not terribly large; massaging would target a big difference, I would think.

-- hide signature --

John

ejmartin Veteran Member • Posts: 6,274
Re: 400D Dynamic range not sensor limited

John Sheehy wrote:

ejmartin wrote:

John Sheehy wrote:

In the shot noise per unit of area, the
FZ50 pixels should outperform all DSLRs at all ISOs.

John, you've stated that your FZ50 performs better in terms of noise
for a given absolute exposure at ISO 100 pushed four stops than it
does at ISO 1600, which is the reverse of any other camera I know of.
It also flies in the face of the notion that the higher ISO
amplification tends to reduce noises when referred back to input
quantities like photoelectrons. I believe you called it a "Panasonic
engineering blooper". Could it be reflective of a problem with the
raw data being massaged by the camera at low ISO?

There are no signs of it; the sigma doubles with each 2x2 binning,
the noise is nice and sharp as you can see.

If so that would
seriously affect your conclusions; for instance, any sort of NR
performed before writing the raw data would throw off the inferred
sensor efficiency, by making it look more efficient than it is. As
it stands, this anomaly in ISO 100 vs 1600 is troublesome, because as
I said higher ISO amplification can only improve input-referred noise
performance.

The proof is in the pudding, though, and that is why I used a real
scene instead of a color checker - there is considerable detail 9 and
10 stops below saturation here; would that be possible with "massaged
noise"? When you look at an ISO 1600 Sony A700 image, the shadows
look all smoothed out. We have nothing like that here.

Oh, and BTW, the difference is not terribly large; massaging would
target a big difference, I would think.

Actually, it looks to my eye that in this comparison,

http://www.pbase.com/jps_photo/image/74342733

there is more and crisper detail in the ISO 1600 shot. And yet you say that ISO 100 has less noise (and indeed it looks that way). But it also looks a bit more smeared; could that be evidence of NR?

Is the gain (electrons/ADU) inferred from raw data proportional to ISO? That I think would be a quantitative test of whether there is some massaging going on in the raw.

-- hide signature --
NIK11 Senior Member • Posts: 2,827
Question for John about new Pana's

John,

I have followed this thread with much interest although I am not too proud to admit that I am beginning to lose the plot. Thanks for all the expert information.

Going by what you have said, or at least what I understand you to have said, and leaving aside all the usual variables and qualifyiers, would you expect the new Pana 10mp LX3 sensor to produce results no better than the 14mp FX150? - both about the same size.

I hope you are a gambling man.

Nick

DSPographer Senior Member • Posts: 2,464
Re: Yes

This is the sort of test I wanted to see. I agree it shows the FZ50 noise to be in the same ball-park per sensor area as the 10D. Of course the 10D design is more than three years older than the FZ50's design so if you get the chance to compare the FZ50 to the 400D with each at their best ISO and pushing the shadows to make the noise obvious then I would certainly like to see that.

In the near future back side illuminated sensors have the potential to outperform front side illuminated sensors in quantum efficiency which means that the smallest sensors may have the highest sensitivity on a per area basis until the larger sensor technology catches up.

 DSPographer's gear list:DSPographer's gear list
Canon PowerShot G7 X Canon EOS 5D Mark II Canon EF 24mm f/2.8 Canon EF 50mm f/1.8 II Canon EF 200mm f/2.8L II USM +4 more
Rriley
Rriley Forum Pro • Posts: 21,846
John

greatly appreciate what you have attempted here

can you explain to me why its better to push - e/v instead of higher iso on small sensors. I have learned on my LC1 that -1 e/v is better than iso200, 400 is out of the question if you seek decent IQ.

what are the implications for dSLRs then, given those circumstances

-- hide signature --

Riley

in my home, the smoke alarm is the dinner bell (just)

 Rriley's gear list:Rriley's gear list
Sigma DP2 Merrill Canon EOS 5D Olympus E-3 Olympus E-5 Panasonic Lumix DMC-GH2 +1 more
ejmartin Veteran Member • Posts: 6,274
Re: Yes

DSPographer wrote:

This is the sort of test I wanted to see. I agree it shows the FZ50
noise to be in the same ball-park per sensor area as the 10D. Of
course the 10D design is more than three years older than the FZ50's
design so if you get the chance to compare the FZ50 to the 400D with
each at their best ISO and pushing the shadows to make the noise
obvious then I would certainly like to see that.

Well, shot noise per unit area was a given. The issue is read noise per unit area, and there we should take the best current DSLR's, say the 1D3 with 4.0 electrons read noise at ISO 1600 and 7.2µ pixels, vs the FZ50 with (according to John; seems a bit low to me) 3.3 electrons read noise at ISO 1600 and 1.97µ pixels. Area scaling just means divide the read noise by the pixel spacing, and so the 1D3 gets 4/7.2~.55, while the FZ50 gets 3.3/1.97~1.7. So the FZ50 will be worse in the read noise dominated regime by over a stop and a half.

-- hide signature --
DSPographer Senior Member • Posts: 2,464
Re: Yes

Shot noise will depend on quantum efficiency so the low shot noise of the FZ50 implies that the sensor manages a good effective fill factor. The backside illuminated sensors will have the possibility of significant improvement in QE versus the current D-SLR.

 DSPographer's gear list:DSPographer's gear list
Canon PowerShot G7 X Canon EOS 5D Mark II Canon EF 24mm f/2.8 Canon EF 50mm f/1.8 II Canon EF 200mm f/2.8L II USM +4 more
ejmartin Veteran Member • Posts: 6,274
Re: Yes

DSPographer wrote:

Shot noise will depend on quantum efficiency so the low shot noise of
the FZ50 implies that the sensor manages a good effective fill
factor. The backside illuminated sensors will have the possibility of
significant improvement in QE versus the current D-SLR.

Shot noise on the FZ50 is competitive with current CMOS DSLR's, neither significantly better nor significantly worse. The discussion was always about read noise.

-- hide signature --
OP John Sheehy Forum Pro • Posts: 21,729
Re: John

Rriley wrote:

greatly appreciate what you have attempted here

can you explain to me why its better to push - e/v instead of higher
iso on small sensors. I have learned on my LC1 that -1 e/v is better
than iso200, 400 is out of the question if you seek decent IQ.

what are the implications for dSLRs then, given those circumstances

I haven't seen this behavior in DSLRs; only in my FZ50. I would assume that the amp used is just garbage. Even when the amp isn't garbage, like in my G9, I tend to use under-exposure for high indices, since high ISOs don't do any special tricks, in any absolute sense. The only reason to use the right ISO or a close one is to see a brighter review image, or get FEC correct. My G9 and FZ50 are usually set to ISO 80/100, and I get up to 360/400 with the EC control. If I want 1600, I'll do it from 400 with -2 EC.

I'd never do this with my Canon DSLRs unless I need the headroom assurance, as their performance is better at high ISOs compared to under-exposures, by a pretty good margin (at least the read noise part, anyway).

-- hide signature --

John

ejmartin Veteran Member • Posts: 6,274
Re: John

John Sheehy wrote:

I haven't seen this behavior in DSLRs; only in my FZ50. I would
assume that the amp used is just garbage. Even when the amp isn't
garbage, like in my G9, I tend to use under-exposure for high
indices, since high ISOs don't do any special tricks, in any absolute
sense. The only reason to use the right ISO or a close one is to see
a brighter review image, or get FEC correct. My G9 and FZ50 are
usually set to ISO 80/100, and I get up to 360/400 with the EC
control. If I want 1600, I'll do it from 400 with -2 EC.

John, what are your measurements on the G9:

Read noise in ADU?
Gain (and at what ISO)?
FWC?

-- hide signature --
OP John Sheehy Forum Pro • Posts: 21,729
Re: A more theoretical visual example

Victor Engel wrote:

http://www.pbase.com/victorengel/image/100405024/original

You may wish to open that link in a separate window and follow along
as I describe it here.

Nice demonstration. It is very ironic that such a powerful example that cuts right to the hollow heart of a myth can result in so little discussion. It almost seems as if, when people keep repeating the so-called "facts" of poor image quality resulting from low-capacity pixels, that they are just in fact doing that; repeating words; with no vision of a picture or model in their heads, of what it is they are talking about. A lot of people probably looked at your demonstrations and shrugged their shoulders, and then went onto another thread and repeated their myths yet again.

Just like what I demonstrated with the original post here; all the nay-sayers keep on saying that small pixels are no good, and when I show that they can actually be better covering the same area, they say "so what?" or make pedantic complaints about things that aren't relevant to the discussion, perhaps not even aware that I have debunked a great part of the myth that they like to repeat.

Anyway, I was thinking about replying to you, and have a new thread related to the subject so I'll link to it in this reply to save an extra post before the 150 limit.

The FZ50 pixels battle the mighty CMOS pixels at ISO 1600 on the 400D:

http://forums.dpreview.com/forums/read.asp?forum=1018&message=28760503

-- hide signature --

John

Keyboard shortcuts:
FForum MMy threads