more pixels are better!

Started Dec 14, 2008 | Discussions
Graystar Veteran Member • Posts: 8,373
Re: more pixels are better!

John Carson wrote:

You presumably haven't seen this:

http://www.dpreview.com/news/0809/08092308fujifilm.asp

I hadn't! Thanks for the link...was interesting.

Well, obviously I can't suggest a 6MP compact to get now! I can't imagine how littered the forum will become with the 6MP vs. 12MP wars this sensor will start. Good thing I don't buy FinePix cameras (for now, at least

Les Olson Senior Member • Posts: 2,081
Re: SNR vs exposure

The weather is indeed lousy, so I have dug out the SNRs at ISO800 for 1% and 10% reflectance from DxO's database. The upshot is that in a multiple regression analysis pixel pitch explains about twice as much of the variation in low light SNR as when the camera was introduced (NB: NOT "when the sensor was designed", which I don't have), but both factors together explain much more than either alone.

The first thing to say is that this analysis is complicated by Nikon's shift to CMOS sensors, which produced a step change (the CMOS cameras have the ):

Camera Date Pixel Pitch SNR, ISO800, 1% SNR, ISO800, 10%

D70 Jan04 7.8 10dB 26dB
D80 Sep06 6.1 8dB 24dB
D40 Nov06 7.8 15dB 27dB
D40x Mar07 6.1 11dB 24dB
D300* Aug07 5.4 14dB 27dB
D3* Aug07 8.4 21dB 34dB
D90* Aug08 5.5 17dB 27dB

When thinking about the effects of pixel pitch vs technical progress, note that the two and a half years the D80 had over the D70 did not compensate for the smaller pixels, but the D40 only two months younger than the D80 but with the same pixel pitch as the D70, has + - two stops less noise in low light than either the D70 or the D80.

Whether better SNR in low light comes from bigger pixels or newer sensor designs, as a rule of thumb the lower the light the bigger the improvement. At 10% reflectance the D80 is "only" one stop worse than the D90, but at 1% it is 3 stops worse, and the D90 is a stop better than the D300 at 1% but the same at 10%. Among Canon designs the 10D (February 03, 7.2 micron pixels) had SNR at ISO800 of 26dB at 10% reflectance and 11dB at 1%, while the 50D (August 08, 4.7 micron pixels) also has SNR of 26dB at ISO800 and 10% reflectance but 15dB at ISO800 and 1% reflectance (more than 1 stop over the 10D).

If low light performance is important to you, is the D40 a bargain or what?

-- hide signature --

2 November 1975.

'... Ma come io possiedo la storia,
essa mi possiede; ne sono illuminato:
ma a che serve la luce?'

Graystar Veteran Member • Posts: 8,373
Re: SNR vs exposure

Les Olson wrote:

Here is a link to the relationship between SNR and exposure:

http://www.microscopyu.com/tutorials/java/digitalimaging/signaltonoise/index.html

Yeah, seen that. Those Nikon MicroscopyU articles are pretty good.

The key point the graphs make is that you need a lot of light to make
sensors photon-noise limited, ie, to make it true that SNR is related
to total sensor area, rather thaan to the size of individual
photosites.

Well, it's John Sheehy's unofficial job to respond to such suppositions, and I'm not one to step on someone else's territory, except to say that there have been several threads in the past with images demonstrating that, when comparing equal image framing in terms of area (e.g. the entire frame of a 1/2.5” sensor vs. a 1/35th crop of a full-frame sensor) when the 1 2.5” sensor is resized to match the lower resolution of the full-frame crop, the noise levels between the two images appears to be the same.

Personally, I suspect that dynamic range depends on pixel size instead of sensor size (i.e. the resized example above still won't exhibit the same DR as the crop.) But there appears to be evidence that seems to support both views. So my own personal jury is still out on that one. But for noise, I'm in agreement with the those in the high-MP camp who say it's sensor size that matters.

bobn2
bobn2 Forum Pro • Posts: 70,895
Re: SNR vs exposure

Les Olson wrote:

Sure, but noise, as a problem that spoils a print or the image on the
screen, is not a highlight problem, it is a shadow problem, and even
where there is a lot of light in a scene overall, the shadows are
still dark. Black is black (I want my detail back) and grey is grey
(I want it here today), what can I do (except turn up the ISOO)?

Well, not completely. Since most output media have a more limited DR than the camera, if you print for the highlights, black will be black.

Plus, we usually take our photographs with a hand-held camera and
often have moving subjects, so we must use short exposures and
cannot, as a scientific or astronomical photographer can, wait for
the wells to fill and therefore use low gain so the black sky stays
black, not speckled.

Nonetheless, most photography situations are comparatively 'lots of light' situations. There are a few people who need a camera which can produce 8+ stops of DR at ISO 25600, but not a lot.

-- hide signature --

Bob

bobn2
bobn2 Forum Pro • Posts: 70,895
Re: SNR vs exposure

Graystar wrote:

Personally, I suspect that dynamic range depends on pixel size
instead of sensor size (i.e. the resized example above still won't
exhibit the same DR as the crop.)

In the end this comes down to science versus superstition. You have your suspicion, but not based on a rigorous process of reasoning or grounded in evaluation of hard evidence. If it were, the conclusion would be that DR ultimately depends on photons collected at an image level, and therefore is dependent on sensor size and efficiency.

But there appears to be evidence
that seems to support both views.

That is wrong, there is no evidence supporting the 'DR depends on pixels size' argument, because it is physically incorrect. What there is is a lot of anecdotes and poorly framed 'tests'. The physics says clearly that DR depends on sensor size and efficiency, I have never seen anyone, anywhere put forward any evidence that shows the contrary. The idea that a concept is valid because a lot of people who don't know what they're talking about say it is, is absurd. Even if some of those people are editors of popular web sites.

So my own personal jury is still
out on that one. But for noise, I'm in agreement with the those in
the high-MP camp who say it's sensor size that matters.

Well done, at least you got to the right conclusion in the end.

-- hide signature --

Bob

ejmartin Veteran Member • Posts: 6,274
Are those figures for "print" or "screen"? (nt)
-- hide signature --
Graystar Veteran Member • Posts: 8,373
Re: SNR vs exposure

bobn2 wrote:

That is wrong, there is no evidence supporting the 'DR depends on
pixels size' argument, because it is physically incorrect. What there
is is a lot of anecdotes and poorly framed 'tests'. The physics says
clearly that DR depends on sensor size and efficiency...

Explain why DxO Labs reports that even though the G10 sensor has a slightly better SNR than the LX3 sensor, the LX3 sensor has one stop more DR.

ejmartin Veteran Member • Posts: 6,274
Re: SNR vs exposure

Graystar wrote:

bobn2 wrote:

That is wrong, there is no evidence supporting the 'DR depends on
pixels size' argument, because it is physically incorrect. What there
is is a lot of anecdotes and poorly framed 'tests'. The physics says
clearly that DR depends on sensor size and efficiency...

Explain why DxO Labs reports that even though the G10 sensor has a
slightly better SNR than the LX3 sensor, the LX3 sensor has one stop
more DR.

There is definitely an error in the ISO calibration of DxO's data for the LX3. My own measurements of G10 vs LX3 at ISO 80 show the following S/N vs exposure:

These are pixel level figures; image S/N involves the S/N at a fixed spatial scale relative to frame height, which involves scaling the pixel S/N ratio by the square root of the MP count. Translated into stops, the G10 curve should be raised by about .25, bringing it closer (but still short of) the LX3 curve. The upshot is that the LX3 has more DR and better S/N throughout the range of exposure levels.

The underlying cause is that the LX3 captures many more photons per unit area than the G10 (.53 e- raw level µ^2 for the LX3 at this ISO; .44 for the G10); and the LX3 has substantially lower read noise (5.6e- per pixel for the LX3; 8.3e- per pixel for the G10).

-- hide signature --
Les Olson Senior Member • Posts: 2,081
Re: Are those figures for "print" or "screen"? (nt)

They are directly measured from the RAW files: http://www.dxomark.com/index.php/eng/Technologies/Measurement-definitions/Noise They are not the "normalised for a 10 micron photosite" or "normalised for a standard print" SNRs used for some of DxO's analyses.
--
2 November 1975.

'... Ma come io possiedo la storia,
essa mi possiede; ne sono illuminato:
ma a che serve la luce?'

Graystar Veteran Member • Posts: 8,373
Re: SNR vs exposure

ejmartin wrote:

There is definitely an error in the ISO calibration of DxO's data for
the LX3. My own measurements...

I see. They're wrong because you're right. Got it.

ejmartin Veteran Member • Posts: 6,274
Re: SNR vs exposure

Graystar wrote:

ejmartin wrote:

There is definitely an error in the ISO calibration of DxO's data for
the LX3. My own measurements...

I see. They're wrong because you're right. Got it.

It's not just that I happen to have done independent quantitative testing (and that I also checked the ISO calibration of the LX3 against a 1D3 and 1Ds3, though not under as controlled a laboratory situation as DxO uses). It's just common sense.

Digicams have 1-3 stops worse S/N performance than DSLR's. If they were to normalize their ISO calibration the same as DSLR's, they would leave about 3-3.5 stops of the DR in highlights. Instead, typical digicams leave about 2.5 stops of headroom, so that they can devote more of what little DR they have to better performance in shadows. This is why it's so much easier to blow highlights on a digicam if you trust the metering, and why many digicam users have found they get better results if they apply at least -1/3 EC to the metering.

Now look at the DxO ISO calibration test for the G10 and LX3 in comparison to say the 1D3:

http://www.dxomark.com/index.php/eng/Image-Quality-Database/Compare-cameras/ (appareil1) 247|0 (appareil2) 188|0 (appareil3) 240|0 (onglet) 0 (brand) Canon (brand2) Canon (brand3) Panasonic

The G10 indeed comes out about 2/3 stop more sensitive than the 1D3; but the LX3 is tested at LESS sensitivity. It makes no sense. And it is not a manufacturer consistent design choice; the Panasonic FZ28 has pretty much exactly the same ISO calibration as the G10.

The combination of my own tests, together with the lack of consistency of DxO's LX3 ISO test with respect to pretty much any other digicam they have tested, leads me to disbelieve their results for the LX3.

-- hide signature --
ejmartin Veteran Member • Posts: 6,274
Re: Are those figures for "print" or "screen"? (nt)

Les Olson wrote:

They are directly measured from the RAW files:

http://www.dxomark.com/index.php/eng/Technologies/Measurement-definitions/Noise They are not the "normalised for a 10 micron photosite" or "normalised for a standard print" SNRs used for some of DxO's analyses.
--

Then you are not comparing apples to apples. SNR is a scale dependent quantity, since signal grows as the square of pixel pitch, and noise grows linearly with pixel pitch, SNR per pixel grows linearly with pixel pitch, for a given level of technology (and with appropriate caveats about read noise). Smaller pixels are destined to have poorer SNR for fixed level of technology. But they also sample a smaller portion of the image. The "normalized for standard print" SNR reported by DxO is the appropriate quantity to compare, because it compares the SNR of a fixed portion of the image.

Equivalently, divide the SNR by the pixel pitch, and compare that. You will find that the D300 and D90 are much better performers, not just for resolution but also for SNR at fixed scale in an image.

-- hide signature --
bobn2
bobn2 Forum Pro • Posts: 70,895
Re: SNR vs exposure

Graystar wrote:

bobn2 wrote:

That is wrong, there is no evidence supporting the 'DR depends on
pixels size' argument, because it is physically incorrect. What there
is is a lot of anecdotes and poorly framed 'tests'. The physics says
clearly that DR depends on sensor size and efficiency...

Explain why DxO Labs reports that even though the G10 sensor has a
slightly better SNR than the LX3 sensor, the LX3 sensor has one stop
more DR.

Not sure how the question relates at all to what I said, but I'll answer it anyway. SNR according to DXO, as I understand it, is the SNR of a 18% grey, exposed as the camera chooses. DR is also essentially an SNR, but the S is the FWC and the N is the read noise. The two will be related some way, but there are a lot of unknown variables in there, like the camera's actual EI for one.

I should clarify my statement and put in the 'all alse being equal'. Of course, you can wreck the DR of an efficient and large sensor by endowing it with very high read noise.
--
Bob

ejmartin Veteran Member • Posts: 6,274
Re: SNR vs exposure

bobn2 wrote:

Graystar wrote:

bobn2 wrote:

That is wrong, there is no evidence supporting the 'DR depends on
pixels size' argument, because it is physically incorrect. What there
is is a lot of anecdotes and poorly framed 'tests'. The physics says
clearly that DR depends on sensor size and efficiency...

Explain why DxO Labs reports that even though the G10 sensor has a
slightly better SNR than the LX3 sensor, the LX3 sensor has one stop
more DR.

Not sure how the question relates at all to what I said, but I'll
answer it anyway. SNR according to DXO, as I understand it, is the
SNR of a 18% grey, exposed as the camera chooses.

Not sure if this is correct. Based on what they are doing, I would think that it's SNR at 18% of raw saturation. They seem rather uninterested in how the camera is metering, and instead simply measuring sensor performance from RAW data.

-- hide signature --
bobn2
bobn2 Forum Pro • Posts: 70,895
Re: SNR vs exposure

ejmartin wrote:

bobn2 wrote:

Graystar wrote:

bobn2 wrote:

That is wrong, there is no evidence supporting the 'DR depends on
pixels size' argument, because it is physically incorrect. What there
is is a lot of anecdotes and poorly framed 'tests'. The physics says
clearly that DR depends on sensor size and efficiency...

Explain why DxO Labs reports that even though the G10 sensor has a
slightly better SNR than the LX3 sensor, the LX3 sensor has one stop
more DR.

Not sure how the question relates at all to what I said, but I'll
answer it anyway. SNR according to DXO, as I understand it, is the
SNR of a 18% grey, exposed as the camera chooses.

Not sure if this is correct. Based on what they are doing, I would
think that it's SNR at 18% of raw saturation. They seem rather
uninterested in how the camera is metering, and instead simply
measuring sensor performance from RAW data.

You are probably right. The reason I said 'as I understand it' is that DxO is very vague on the detail of its test methods, one of the reasons its less useful than it might be.

-- hide signature --

Bob

Les Olson Senior Member • Posts: 2,081
Re: Are those figures for "print" or "screen"? (nt)

I disagree. For any quantity the correct approach is to compare first the actual value. Adjusted, normalised, or standardised values can be compared IF they serve a purpose, and it does not matter whether you are talking about "seasonally adjusted unemployment" or normalised SNR. The trap, as in this case, is to use normalisations that encode ideas about the right answer.
--
2 November 1975.

'... Ma come io possiedo la storia,
essa mi possiede; ne sono illuminato:
ma a che serve la luce?'

Graystar Veteran Member • Posts: 8,373
Re: SNR vs exposure

bobn2 wrote:

Graystar wrote:

Explain why DxO Labs reports that even though the G10 sensor has a
slightly better SNR than the LX3 sensor, the LX3 sensor has one stop
more DR.

Not sure how the question relates at all to what I said

You said that DR depends on sensor size and efficiency. The two sensors are roughly the same size (LX3 sensor is technically a tiny bit bigger, but is never entirely used) and the very close SNR would implicate similar efficiency. Yet the DR is very different...one stop is quite a bit. From your point of view the DR should be the same.

That's how it relates to what you said.

bobn2
bobn2 Forum Pro • Posts: 70,895
Re: SNR vs exposure

Graystar wrote:

bobn2 wrote:

Graystar wrote:

Explain why DxO Labs reports that even though the G10 sensor has a
slightly better SNR than the LX3 sensor, the LX3 sensor has one stop
more DR.

Not sure how the question relates at all to what I said

You said that DR depends on sensor size and efficiency. The two
sensors are roughly the same size (LX3 sensor is technically a tiny
bit bigger, but is never entirely used) and the very close SNR would
implicate similar efficiency. Yet the DR is very different...one
stop is quite a bit. From your point of view the DR should be the
same.

That's how it relates to what you said.

I didn't say depends solely on sensor size and efficiency - that would be silly. The floor of the DR is the read noise. Double that between cameras (and it varies much more than that) and you have your stop in DR. Sorry for not being more precise, I was trying to counterpose sensor size/efficiency against pixel density, as I said the read noise forms the floor, and is the major of the 'all other things being equal' which needs to go into any such statement. Also, the variation in read noise (and for that matter FWC) means that SNR is not a direct indicator of efficiency.

-- hide signature --

Bob

bobn2
bobn2 Forum Pro • Posts: 70,895
Re: Are those figures for "print" or "screen"? (nt)

Les Olson wrote:

I disagree. For any quantity the correct approach is to compare
first the actual value.

And the 'actual value' is the noise/DR etc per unit area of sensor, right?

-- hide signature --

Bob

Graystar Veteran Member • Posts: 8,373
Re: SNR vs exposure

bobn2 wrote:

I didn't say depends solely on sensor size and efficiency - that
would be silly. The floor of the DR is the read noise. Double that
between cameras (and it varies much more than that) and you have your
stop in DR.

So you're saying what...that the LX3 doesn't do any better in highlights than the G10? It's all shadow detail?

Keyboard shortcuts:
FForum MMy threads