Digital Camera Sensor Performane Summary updated Feb 3

I'll try and clear up some of the confusion spewed in the
last few posts.

Ilya Zakharevich wrote:
> ??? I was not discussing any "trend". What I did was I compared 1DmII
> with 1DmIII, and noticed that the performance at 50ISO is 90% compatible
> with an assumption that they use the same ADC, and just the
> capacitance changed.

There are two (at least) amplification steps to get the signal
off of an electronic sensor. 1) the read (also called sense) amplifier,
which is a fixed gain (does not change with ISO). This is the
source of the sensor read noise. 2) Second amplifier, which
may be integrated with the ADC. The amplifier changes gain
with ISO.

Sensor read noise comes from the first amplification
stage off the pixel/chip. It does not change with ISO.
Ilya says the the lower (read) noise seen in the low ISO 1DIII
is due to the capacitance of the smaller pixel. But the
noise at low ISO is dominated by the second stage amplifier
(the ADC system) not the read amplifier. Typical noise levels from
this second stage amplification is 6 to 8 times the read noise
from the first stage amplifier. A model for the two amplifiers
and observed data are shows in Figure 8 at:
http://www.clarkvision.com/imagedetail/digital.sensor.performance.summary

So when you want to look at true sensor read noise, low ISOs is
not the way to see it. Ilya's assertion of capacitance controlling
low ISO noise is simply incorrect.

To see the real sensor read noise, the secondary system noise
must be minimized. In current digital cameras, the only
way the have sensor read noise dominate is at high ISO.

>> "In the physics of photon counting, the noise in the signal is equal
>> to the square root of the number of photons counted"
>
Ilya Zakharevich wrote:
> Wrong. This is "photon"/electron noise; one of three major components
> of noise for short exposures (electron Poisson noise, pre-amplifier
> read noise, post-amplifier read noise). And in the quoted part, we
> were discussing read noise for small ISO (i.e., post-amplifier read
> noise).

This is pretty simple math. The noise from modern digital
cameras is dominated by:

N = sqrt(P + r*r + t*t),

Where N = total noise in electrons, P = number of photons,
r = read noise in electrons, and t = thermal noise in electrons.

t < 1 electron/second so for short exposures can be ignored.

P maxes out at ~80,000 photons in large pixel cameras at low ISO,
and is even around 50,000 photons on cameras like the Canon 30D.
Read noise is typically around 4 electrons at high ISO, but at
the low ISO of the max sensor signal, is around 25 to 50 electrons
at low ISOs (from the second stage amplifier+ADC). So where is
the signal photon noise limited?

Example, Canon 20D, 30D at ISO 100:

Signal level Photon Noise read noise (+ADC noise)
Photons Stops
50,000 0 223 25
25,000 -1 158 25
12,500 -2 119 25
6.250 -3 79 25
3,125 -4 56 25
1,562 -5 40 25
781 -6 28 25
391 -7 20 25

So photon noise is greater than read noise for more than 6
stops down.

Example, Canon 20D, 30D at ISO 800:

Signal level Photon Noise read noise (+ADC noise)
Photons Stops
6,520 0 70 4.8
3,125 -1 56 4.8
1,562 -2 40 4.8
781 -3 28 4.8
391 -4 20 4.8
195 -5 14 4.8
98 -6 10 4.8
49 -7 7 4.8
24 -8 4.9 4.8
12 -9 3.5 4.8

So at ISO 800, photon noise is larger than read noise for
more than 8 stops on the 20D and 30D.

Summary: Modern digital cameras have noise that is photon noise dominated
(called photon noise limited) of a large range of intensities.
The 14-bit converters are pushing that second stage noise source
lower so they are photon noise limited over a larger range.

You'll see read noise (and fixed pattern noise) mainly in the deepest
shadows of your digital camera images, but not in the mid-tones
and highlights.

Roger
 >> Stay informed about: Digital Camera Sensor Performane Summary updated Feb 3 

On Feb 13, 1:49 am, alex <a....TakeThisOut@spam.me.not> wrote:
> Roger N. Clark (change username to rnclark) wrote:
>
> > Count me as one who will rarely, if ever participate in the future.
>
> That is a pity. I am sure a lot of people read this NG and value your
> comments (I certainly do). I think in NGs where idiots are few and far
> between (a-la JSH on sci.math or spinoza1111 in comp.programming),
> it makes sense to argue with them (or at least flag their posts).
>
> When they are more numerous I just killfile them. Tune them out
> even if 20 of them run in and scream "what a stupid thing this is",
> "yes! how stupid!" after each of your posts. I

Yes but Roger doesn't listen. If you look at his last few posts (over
the last few months), most of them are arguing with said idiots, even
after it becomes blatantly obvious that it's pointless. Predictably
enough, this got old (in combination with all sorts of personal
attacks on him, of course).

He's not the only one like this, many people have left in this way.
Too bad.
 >> Stay informed about: Digital Camera Sensor Performane Summary updated Feb 3 

alex wrote:
> Roger N. Clark (change username to rnclark) wrote:
>> Count me as one who will rarely, if ever participate in the future.
>
> That is a pity. I am sure a lot of people read this NG and value your
> comments (I certainly do). I think in NGs where idiots are few and far
> between (a-la JSH on sci.math or spinoza1111 in comp.programming),
> it makes sense to argue with them (or at least flag their posts).
>
> When they are more numerous I just killfile them. Tune them out

Agreed!
The list is predictable and others will clarify if you simply ignore.

> even if 20 of them run in and scream "what a stupid thing this is",
> "yes! how stupid!" after each of your posts. It is not much more
> difficult to carry a conversation in this environment, just choose
> whom to respond to -- this is not a seminar where you feel obliged
> to respond to every comment.
>
> Have fun!
>
 >> Stay informed about: Digital Camera Sensor Performane Summary updated Feb 3 

On Tue, 12 Feb 2008 23:30:01 -0700, "Roger N. Clark (change username to
rnclark)" <username RemoveThis @qwest.net> wrote in <47B28E69.3080608 RemoveThis @qwest.net>:

>I'll try and clear up some of the confusion spewed in the
>last few posts.

>...

>So when you want to look at true sensor read noise, low ISOs is
>not the way to see it. Ilya's assertion of capacitance controlling
>low ISO noise is simply incorrect.

That may well be true, but (and with all due respect) you haven't proved
that -- you've just made a number of assertions.

--
Best regards,
John Navas
Panasonic DMC-FZ8 (and several others)
 >> Stay informed about: Digital Camera Sensor Performane Summary updated Feb 3 

[A complimentary Cc of this posting was sent to
Roger N. Clark (change username to rnclark)
<username RemoveThis @qwest.net>], who wrote in article <47B28E69.3080608 RemoveThis @qwest.net>:
> There are two (at least) amplification steps to get the signal
> off of an electronic sensor. 1) the read (also called sense) amplifier,
> which is a fixed gain (does not change with ISO). This is the
> source of the sensor read noise. 2) Second amplifier, which
> may be integrated with the ADC. The amplifier changes gain
> with ISO.
>
> Sensor read noise comes from the first amplification
> stage off the pixel/chip. It does not change with ISO.
> Ilya says the the lower (read) noise seen in the low ISO 1DIII
> is due to the capacitance of the smaller pixel.

What I said was close to this:

a) The area of newer sensel is about 77% of the old one ((7.2/8.2)**2);

b) The low-ISO read noise of the newer sensor+ADC (in electrons) is
about 79% of the old one (24.4/30.6).

Assuming capacitance change proportional to the area change, and
taking into account V = Charge/Capacitance, this means that

*) The low-ISO read noise of the newer amplifiers+ADC (in volts) is
about 103% of the old one.

My statement does not depend on any particular detail of how the
measurement pipeline is designed.

> But the noise at low ISO is dominated by the second stage amplifier
> (the ADC system) not the read amplifier.

The details of the pipeline are irrelevant to what I wrote. What a
change to 14-bit output could improve was performance at low-ISO (one
does not need more than 9-10bits for high ISO). It did not.

[I had no chance to read the rest, so I comment only on this part of
your post.]

Thanks for your clarifications anyway,
Ilya
 >> Stay informed about: Digital Camera Sensor Performane Summary updated Feb 3 

[A complimentary Cc of this posting was sent to
Roger N. Clark (change username to rnclark)
<username.TakeThisOut@qwest.net>], who wrote in article <47B28E69.3080608.TakeThisOut@qwest.net>:
> Summary: Modern digital cameras have noise that is photon noise dominated
> (called photon noise limited) of a large range of intensities.

.... and the SIZE of this range depends on the OTHER components of
noise (mainly, the two components of the read noise).

> The 14-bit converters are pushing that second stage noise source
> lower so they are photon noise limited over a larger range.

Higher-bits converters do not push anything, since the only situation
the higher bit count matters is low-ISO, and they have EXACTLY the
same low ISO performance.

What the 1DmIII has (in MAJOR advantage to 1DmII) is significantly
lowered pre-amplifier noise (about 1.5x lower [calculate as
(7.2/8.2)**2/(2.07/4)]). But low-preamplifier noise is not sensitive
to the bit count of ADC; as I have shown earlier, a slightly better
than 9-real-bits ADC will not worsen the errors.

> You'll see read noise (and fixed pattern noise) mainly in the deepest
> shadows of your digital camera images, but not in the mid-tones
> and highlights.

Sure. And to decrease these (i.e., increase useful dynamic range,
possibly by decreasing resolution in these areas), one needs low read
noise.

Hope this helps,
Ilya
 >> Stay informed about: Digital Camera Sensor Performane Summary updated Feb 3 

"Roger N. Clark (change username to rnclark)" <username.DeleteThis@qwest.net> wrote:

> I'll try and clear up some of the confusion spewed in the
> last few posts.

Which your very clear explanation certainly did for me.

Thanks!

--
Chris Malcolm cam.DeleteThis@infirmatics.ed.ac.uk DoD #205
IPAB, Informatics, JCMB, King's Buildings, Edinburgh, EH9 3JZ, UK
[http://www.dai.ed.ac.uk/homes/cam/]
 >> Stay informed about: Digital Camera Sensor Performane Summary updated Feb 3 

On Feb 6, 11:11 pm, "Roger N. Clark (change username to rnclark)"
<usern....RemoveThis@qwest.net> wrote:
>
> Look at higher ISOs, and the 1DIII read noise drops to
> 2.1 electrons compared to the 1DII 3.9 electrons, about
> an 86% improvement, which helps high ISO imaging.
>
>
> "Hope this Helps"
> Roger

Roger, I've been wondering in comparing my own measurements of the 1D3
whether there is a clerical error in your 1D3 data. My data on a 1D3
gave 5.1 electrons/14-bit ADU at ISO 100, while you report a number
about half that (2.46). Correspondingly, the high ISO read noise
bottoms out around 2 electrons in your data, whereas I have it at
about 4 electrons (which would be explained if the gain measurement
were off by two). If your gain figure is accurate, the 1D3 is a big
step back from the 1Ds2 which has the same size pixels and 3.3
electrons/12-bit ADU at ISO 400 (and so 3.3 in 14-bit units at ISO
100). In fact, it would perform even more poorly than the 20D. I
suspect that the MkIII data in the first column of table 3b in your
summary page needs to be shifted down by one row for each entry, then
it makes more sense and is compatible with what I am seeing.
 >> Stay informed about: Digital Camera Sensor Performane Summary updated Feb 3 

On Feb 4, 12:29 am, "Roger N. Clark (change username to rnclark)"
<usern... DeleteThis @qwest.net> wrote:
> If you haven't visited this month, see:
>
> http://www.clarkvision.com/imagedetail/digital.sensor.performance.sum...
>
> I've added data for several newer cameras, added a new
> section on Apparent Image Quality (AIQ) along with models
> for constant megapixels and constant format size. What formerly
> looks like a scatter plot now is taking shape in that the
> pixel size and sensor format are showing trends. The models
> should now allow prediction of performance when new cameras
> come out. The new AIQ plot is Figure 9.
>
> Roger

I have some puzzles and comments about the notion of AIQ. The way I
read it, apart from some constants that drop out of any comparisons
among cameras, it is defined as AIQ = const. x sqrt(full well
electrons) x megapixels. This is an attempt to combine distinct two
measures of image quality (SNR and resolution), and one can ask why
they should be combined in exactly this way, but let's accept the
definition as it stands. Resolution in line widths per picture
height, assuming that the resolution goes out to near the Nyquist
frequency, is given by the sqrt of the pixel count up to a universal
constant. Then AIQ, up to constants, is (SNR)x(linear
resolution)^2.

I am confused about the further analysis of the effect of diffraction
limitation. Surely diffraction limits resolution, but it does not
affect SNR since the same photons are being collected, they are merely
being redistributed among the pixels due to the diffraction
phenomenon. So when evaluating SNR, we should do it over the area of
the diffraction spot size, combining the pixels that are within the
the dominant support of the diffraction spot. The signal will scale
with the number of pixels, the noise with the sqrt of the number of
pixels, and so the SNR of the diffraction spot will scale with the
sqrt of the number of pixels within it. So really we should think of
the factors in diffraction limited AIQ as

AIQ = const. x (SNR in a diffraction spot) x (diffraction limited
resolution)^2

I am wondering whether you accounted for this in your analysis, since
it is not the SNR of a pixel that should enter, but rather the SNR of
a diffraction spot, and this scales as the SNR of a pixel times the
size of the diffraction spot in pixel widths.

I would also think that a slightly different measure along these lines
would be more appropriate. When one looks at an image one takes it in
as a whole, and looks at features within the image. The SNR compenent
of AIQ should then be measured with respect to the fixed size features
that are being rendered by the camera, in other words one should look
at SNR per unit area and not the SNR of a pixel. Two slightly
different versions are possible -- either look at SNR per unit sensor
area, or SNR as a percentage of the area of the frame (depending on
how one wants to treat crop factor). Then for the same reasons as
above, the SNR per area scales with the sqrt of the megapixel count.
We then write

AIQ = const. x sqrt(full well electrons x megapixels) x
sqrt(megapixels)
= const. x (SNR per area) x (linear resolution)

Where "megapixels" is either megapixels per area, or total megapixels
depending on how one wants to think about crop factor. Now when
diffraction limitation arises, the first factor is unchanged, and the
second factor is replaced by the diffraction limited linear resolution
(in either line pairs per mm or line pairs per picture height,
depending on how one wants to treat crop factor).

cheers,
emil
 >> Stay informed about: Digital Camera Sensor Performane Summary updated Feb 3 

On Feb 15, 9:59 am, ejmartin <ejm_60... RemoveThis @yahoo.com> wrote:
> On Feb 6, 11:11 pm, "Roger N. Clark (change username to rnclark)"
>
> <usern... RemoveThis @qwest.net> wrote:
>
> > Look at higher ISOs, and the 1DIII read noise drops to
> > 2.1 electrons compared to the 1DII 3.9 electrons, about
> > an 86% improvement, which helps high ISO imaging.
>
> > "Hope this Helps"
> > Roger
>
> Roger, I've been wondering in comparing my own measurements of the 1D3
> whether there is a clerical error in your 1D3 data. My data on a 1D3
> gave 5.1 electrons/14-bit ADU at ISO 100, while you report a number
> about half that (2.46). Correspondingly, the high ISO read noise
> bottoms out around 2 electrons in your data, whereas I have it at
> about 4 electrons (which would be explained if the gain measurement
> were off by two). If your gain figure is accurate, the 1D3 is a big
> step back from the 1Ds2 which has the same size pixels and 3.3
> electrons/12-bit ADU at ISO 400 (and so 3.3 in 14-bit units at ISO
> 100). In fact, it would perform even more poorly than the 20D. I
> suspect that the MkIII data in the first column of table 3b in your
> summary page needs to be shifted down by one row for each entry, then
> it makes more sense and is compatible with what I am seeing.

Just to make this point more emphatically -- taking the data in tables
3a,b and 4a,b literally, the 1D3 has the worst gain performance of any
Canon DSLR listed. Simultaneously, the high ISO read noise is almost
a factor of two better than all other Canon DSLR's listed.

I suspect you don't really mean that

All would be well if the gain values were twice what is quoted for
each ISO, then that propagates through to the read noise values which
will be around 4 electrons at high ISO which is more in line with all
other Canons in your tables. I don't think the DR should be affected
since the gain cancels out of that between electrons at saturation and
read noise in electrons.

On the dark side, I'm puzzled as to some of the D300 values quoted
(which I think were based on my analyses). The gain figures are OK,
but the read noises should be

ISO Read noise (12bit/14bit)
200 7.2/6.6
400 6.2/5.9
800 5.2/5.1
1600 4.9/4.6
3200 4.8/4.7

according to my data. None of these numbers seem to bear much
relation to what is listed in your table 4b.

BTW, if you're interested in a more civilized venue for technical
discussions, I might suggest openphotographyforums.com.
 >> Stay informed about: Digital Camera Sensor Performane Summary updated Feb 3 

On Feb 15, 2:01 pm, Ilya Zakharevich <nospam-ab....TakeThisOut@ilyaz.org> wrote:
> [A complimentary Cc of this posting was sent to
> ejmartin
> <ejm_60....TakeThisOut@yahoo.com>], who wrote in article <63ba6203-0e09-4e1f-8717-95eaa5af0....TakeThisOut@u10g2000prn.googlegroups.com>:
>
> > among cameras, it is defined as AIQ = const. x sqrt(full well
> > electrons) x megapixels. This is an attempt to combine distinct two
> > measures of image quality (SNR and resolution), and one can ask why
> > they should be combined in exactly this way
>
> Of course, this measure is just what Rodger's left pinky wanted.
> However, note that until some reasonable investigations of psychology
> of vision are performed/found, we are confined to guesswork anyway.
> So I, personally, do not have a large problem with this particular
> part of what Roger added.

Well, I would have said that two a priori independent quantities
should be reported as independent quantities rather than multiplied
together. But that's just me.

>
> > I am confused about the further analysis of the effect of diffraction
> > limitation. Surely diffraction limits resolution, but it does not
> > affect SNR since the same photons are being collected, they are merely
> > being redistributed among the pixels due to the diffraction
> > phenomenon. So when evaluating SNR, we should do it over the area of
> > the diffraction spot size, combining the pixels that are within the
> > the dominant support of the diffraction spot.
>
> Did not you see my posts about relation of dynamic range and
> resolution?

Um, no. I left this place for a long time because the SNR got too
low.

> The point is that given DSP, it does not make sense to
> consider them separably. Another point is that considering signal and
> noise in the frequency domain (i.e., after Fourier transform) makes
> things much more clear.
>
> The coming signal is filtered by the MTF of the system, and some noise
> is added to it. The intensity of this noise also depends on the
> spacial frequency; so you get two curves of intensity vs spacial
> frequency: the signal, and the noise.
>
[snip]
>
> IMO, an "artistically useful" signal is one with S/N >= 3. So one
> would "cut out" the curves at the point where S/N=3; this gives yet
> another curve, which has the meaning of the maximal resolution vs the
> zone. This "dynamic range curve" should replace the (IMO, much less
> useful) notion of the "dynamic range number".

Well, yes, all valid information but probably not helpful. The art of
data interpretation is to take the mass of data and distill from it a
small number of parameters that distill the essence of their meaning.
Just listing all the data in a repackaged form is not particularly
useful. Going in the other direction, trying to boil it all down to
one magic number, as in Roger's AIQ, may not be meaningful either. In
particular there are going to be degeneracies -- is a camera with good
SNR and poor resolution better or worse than one with poor SNR and
good resolution? That depends on individual preferences regarding
noise and resolution; I know some people who just abhor noise, to the
point that they smear fine scale detail to pot in order to get rid of
it, and others who do little or no NR in order to maximize detail.

I think I can keep straight a few figures of merit, not too many and
not too few. And they should be measured on a level playing field --
for instance, DR and midtone SNR should be measured at fixed spatial
frequency (so on that part we are in agreement), not at the pixel
level which is a moving target as pixel sizes change from model to
model, meaning that these quantities are being measured at different
spatial frequencies and then compared, which I find a bit silly. That
is why I suggested that midtone SNR should be scaled by pixel pitch to
get a more appropriate measure of image SNR.
 >> Stay informed about: Digital Camera Sensor Performane Summary updated Feb 3 

[A complimentary Cc of this posting was sent to
Roger N. Clark (change username to rnclark)
<username RemoveThis @qwest.net>], who wrote in article <47A6B0B2.8040005 RemoveThis @qwest.net>:
> If you haven't visited this month, see:
>
> http://www.clarkvision.com/imagedetail/digital.sensor.performance.summary
>
> I've added data for several newer cameras, added a new
> section on Apparent Image Quality (AIQ) along with models
> for constant megapixels and constant format size.

Thanks again for the numbers; I've sat over the following remarks for
a week now, and I think they are ripe for publication now :

Yours,
Ilya

=======================================================

I think the idea to have something like Figure 9 is very useful. However,
as it often happens with the first draft, the usefulness of the current
version is negligible...

a) The QE numbers used in the document are half-an-order-of magnitude off.
(The reason is that the losses due the Bayer filtering are not taken into
account.)

b) The lines for constant-MP-count do not have any relation to reality.
Smaller-die sensors can use technologies which produce higher defect
densities without endangering the yield. Thus using the same value
1um for inter-sensels gap across different die sizes is not very
meaningful. (E.g., a typical small sensor of today has a step of
1.8um; this gives the "expected" gap about 0.5um - to get fill-factor
about 50%.)

c) Inter-sensel gap should decrease as technology improves (about 20-30%
a year). As color-separation technology improves, the QE would improve
as well. Thus the picture reflects "state-of-art of 2007"; it would be
honest to mark it as such (some people might think these are theoretical
estimates with the BEST POSSIBLE technology, not the technology of 2007).

d) The semantic of the metric for quality (Sqrt(Full_Well) * MPix) has
relation to the performance at very bright light. It would be nice if
the diagram would be marked as "very bright light performance" or somesuch.

e) It does not make sense to couple a larger sensor with a brighter lens.
The larger the sensor, the harder it is to make a decently-performing-
at-the-same-f/stop lens. Thus using f/8 with 1/6x-crop sensor, while
using f/4 with a full-frame sensor does not make a lot of sense.

f) Since f-stop does not affect full-well or MPix count, I do not see how
f-stop curves sneaked into the diagram.

g) THERE IS some relation between abberation of the lens, and a useful MPix
count. But the relation implied by the diagram is almost an-order-of-
magnitude off the reality.

As shown, e.g., in
http://groups.google.com/group/rec.photo.digital/browse_thread/thread/...ca470ad
with f/8 diffration-limited lens one should get significant improvements
in the image quality up to about 120MP full-frame sensor; about this value
the improvements start to saturate.
 >> Stay informed about: Digital Camera Sensor Performane Summary updated Feb 3 

[A complimentary Cc of this posting was sent to
ejmartin
<ejm_60657 RemoveThis @yahoo.com>], who wrote in article <63ba6203-0e09-4e1f-8717-95eaa5af0519 RemoveThis @u10g2000prn.googlegroups.com>:
> among cameras, it is defined as AIQ = const. x sqrt(full well
> electrons) x megapixels. This is an attempt to combine distinct two
> measures of image quality (SNR and resolution), and one can ask why
> they should be combined in exactly this way

Of course, this measure is just what Rodger's left pinky wanted.
However, note that until some reasonable investigations of psychology
of vision are performed/found, we are confined to guesswork anyway.
So I, personally, do not have a large problem with this particular
part of what Roger added.

> I am confused about the further analysis of the effect of diffraction
> limitation. Surely diffraction limits resolution, but it does not
> affect SNR since the same photons are being collected, they are merely
> being redistributed among the pixels due to the diffraction
> phenomenon. So when evaluating SNR, we should do it over the area of
> the diffraction spot size, combining the pixels that are within the
> the dominant support of the diffraction spot.

Did not you see my posts about relation of dynamic range and
resolution? The point is that given DSP, it does not make sense to
consider them separably. Another point is that considering signal and
noise in the frequency domain (i.e., after Fourier transform) makes
things much more clear.

The coming signal is filtered by the MTF of the system, and some noise
is added to it. The intensity of this noise also depends on the
spacial frequency; so you get two curves of intensity vs spacial
frequency: the signal, and the noise.

By doing DSP, we can move these curves up and down arbitrarily; so the
only data which is invariant wrt postprocessing is the quotient, which
is the curve of S/N ratio vs spacial frequency. Since noise depends
on the light intensity, this curve depends on the zone as well. So
what you get is a family of curves, (e.g.) one per zone.

IMO, an "artistically useful" signal is one with S/N >= 3. So one
would "cut out" the curves at the point where S/N=3; this gives yet
another curve, which has the meaning of the maximal resolution vs the
zone. This "dynamic range curve" should replace the (IMO, much less
useful) notion of the "dynamic range number".

So far, we did not use the S/N-vs-frequency curves much, only used one
point in each curve. With my minimal experiments, another useful
metric one could extract from these curves is the information content
(calculated by Shannon): the area below this curve drawn in the
log-scale (with cutout at S/N=3). AFAIK, this number correlates well
with the "visual quality" of the image.

[This number depends on the zone too. So one gets another curve which
shows how much the information density is in the image at each zone.]

Hope this helps,
Ilya
 >> Stay informed about: Digital Camera Sensor Performane Summary updated Feb 3