[kh1234567890]

Then also there are mitigating effects.

Oh no, not the myth that downsampling reduces noise. It will do, but only on a linear encoded image - not on a gamma encoded JPEG.

[ElJamoquio]

You probably mean: "Equivalence is dumb. It uses math in a dumb way."

Equivalence is dumb, it shows me when advertisers have misled me into purchasing the wrong product.

[wombat2go]

JSH, Yes I follow most of your post #204 .
But the "Total light" has nothing to do with a bigger sensor frame being a better performer in terms of signal to noise ratio and resultant dymamic range and hence that concept can't support the title of this thread.

For example if Total Light is a figure of merit, we could take an imaginary 4 sensors from an ist d glue them together, put on a lens with same FOV and then see less noise at iso 1600 from the 4 glued sensors compared a normal ist d. Of course we know by intuition that can't be true.
If that were true there would be an enormous difference in signal to noise ratio between a Pentax camera and a pro-grade digital back, and (I haven't checked lately) they are not so different are they?.

You have to consider luminous flux (lumen per steradian) on the sensor well. That sets the forward bias on each photodiode and hence its signal to noise ratio.
And the signal to noise ratio of each well varies across the image, being high on bright and low on dark.
Each well has a particular signal to noise ratio unless the image is a uniform flat plane and Cosfourth of the lens is ignored.

A "sensor wide" average of dynamic range based signal to noise ratio of "Total light" from an image would be meaningless in my opinion.
Is such a figure of merit being used by those in your reference? If so it would have to be qualified by sensor size, because it will increase by sensor size.

The luminous flux level we set for the sensor is based on setting the f/- and shutter to the ISO rating of the sensor and is unrelated to the sensor size.
Again we know that by common sense, I have 5 format sizes here from m4/3 to 4x5, and the same light meter can set tham all.

The digital manufacturers did not have to continue with the {ISO, F/- and time} But as far as I have read, all manufacturers of consumer and pro grade cameras did. They did not introduce exposure control based on "Total Light" and sensor size, for the above reason.

The reason bigger sensors can be better is to do with their more modern technology, and the Gassian composite spread function can be smaller, and it does not have to be magnified so much on am image sized for our screens or prints.
In my opinion none of that is related to the issue of this thread title.

[jsherman999]

Oh no, not the myth that downsampling reduces noise. It will do, but only on a linear encoded image - not on a gamma encoded JPEG.

Pages and pages of debate on this issue, on how it reduces 'noise', if it even really does, and if it does how much, and how inneficient it is as a method vs. as simply a by-product compared to actual noise reduction algorithms.

Based on my experience with my D800 and lesser-MP cameras, and what I see at different display sizes when using no NR, and what I see in the measured results form the 24mp vs. the 36mp sensors.... I'm coming down on the side of DXOmark and the others who say it's not a myth

You are free to disagree of course.

---------- Post added 10-11-14 at 11:35 AM ----------

JSH, Yes I follow most of your post #204 .
But the "Total light" has nothing to do with a bigger sensor frame being a better performer in terms of signal to noise ratio and resultant dymamic range and hence that concept can't support the title of this thread.

For example if Total Light is a figure of merit, we could take an imaginary 4 sensors from an ist d glue them together, put on a lens with same FOV and then see less noise at iso 1600 from the 4 glued sensors compared a normal ist d. Of course we know by intuition that can't be true.

Actually unless I misunderstand what you're saying it would be true if you maintained the FOV while using all four of those glued-together sensors to make the image. In other words, if you a big enough lens + used a different FL, which would need to result in a greater physical aperture used for the same exposure. It's not going to be a linear increase because I don't think the underlying *istD tech would support it, but your frankenstein sensor there would conceivably perform somewhere between the original *stD and... something bigger - given a matching lens.

The main reason it's not just 'make it so then' is because of yields and wafer cost such a sensor would be enormously expensive, not to mention that it would need a whole new mount and line of lenses. I don't think we're going bigger than cropped medium format anytime soon because of those economics and what the market can support.

If that were true there would be an enormous difference in signal to noise ratio between a Pentax camera and a pro-grade digital back, and (I haven't checked lately) they are not so different are they?.

The main limitation there is the lenses' physical apertures, and the sensor tech itself. I think most 'fast' MF lenses are around f/2.8. Add to that the sensor efficiency typically is lower as you move up - 645D case in point, a fairly low-efficiency MF CCD sensor.

Also, the difference in sensor area between say the 645D/Z and FF is smaller than the delta between aps-c and FF, so we probably wouldn't expect to see the same jump in performance anyway, even with equally efficient sensors. We'll see more when we see the 645Z numbers.

And the signal to noise ratio of each well varies across the image, being high on bright and low on dark.
Each well has a particular signal to noise ratio unless the image is a uniform flat plane and Cosfourth of the lens is ignored.

That very point was actually brought up in response to The_Suede's post in that thread, I think you should really read through the thread as it's addressed. Basically what you say is true, but is true for any sensor size and isn't fully pertinent to the discussion of total light for the image. (please read the thread carefully, I don't think we should be translating and pasting here, especially when the original contributors are living/breathing people who can be queried directly with a PM.)

The luminous flux level we set for the sensor is based on setting the f/- and shutter to the ISO rating of the sensor and is unrelated to the sensor size.

Yes, we've gone over this That's the exposure. Not the radiant flux, not the total light. We have no disagreement re exposure.

.

Again we know that by common sense, I have 5 format sizes here from m4/3 to 4x5, and the same light meter can set tham all.

Light meter judges exposure values.

The reason bigger sensors can be better is to do with their more modern technology,.

No - really not the Case. The ancient Canon 5D sensor performs slightly better than the more modern K-5/Sony (in resulting image noise,) and the almost equally ancient D700 still performs about a stop better. In those cases the more modern technology of the aps-c sensor is mitigating the effects of total light - in the case of the much less efficient Canon, the mitigation is much more successful because the sensor is that much worse.

---

Last edited by jsherman999; 10-12-2014 at 10:45 AM.

[noser]

The ancient Canon 5D sensor performs slightly better than the more modern K-5/Sony (in resulting image noise,) and the almost equally ancient D700 still performs about a stop better. In those cases the more modern technology of the aps-c sensor is mitigating the effects of total light - in the case of the much less efficient Canon, the mitigation is much less successful because the sensor is that much worse.

The 5D pixel pitch is 8 um, the K-5 is 4.8 um. The modern tech at work here in the K-5 is attempting to level the playing field of having only a third of the light gathering area per pixel (and the resulting s/n ratio) to work with. And it's seems to almost balance things out. Almost...

Is that what you're saying? 'Cause that's what's happening. [edit: I think I'm agreeing with your disagreement in your post... bigger != newer]

When I started reading this thread I did a total eye-roll / mouth-barf, but I decided to keep reading for the entertainment and limited science vs myth value... If only Jamie and Adam were here; they'd blow something up...

[Aku Ankka]

You don't question exposure but you keep pushing total light as if it is independent from total sensor area. You keep claiming that as long as total captured light is the same then noise will be the same.

Assuming that the sensor related noises (which are minute and irrelevant outside of deep shadows) are the same, then he is right. The amount of noise (ie. standad deviation) in light itself is equal to the square root of the number of photons. 10000 photons captured means standard deviation of 100 and SNR of 100.

Photography has always been total light over total area.

Didn't you just disagree with that?

Smaller sensors need lesser light. Exposure is maintained. Noise is maintained. It's very basic.

While the exposure is maintained, the effect of exposure will be different. An f/2.8 image on an iPhone will have vastly lower signal-to-noise ration than an f/2.8 image on a medium format camera.

If you disagree what I wrote in bold, would you explain why it would be wrong.

If you want to have the same SNR on a smaller sensor, you will either need to increase the aperture (ie. reduce the aperture number), or increase the exposure time. Just like you said: it's abut the total light.

---------- Post added 18-10-14 at 14:12 ----------

AoV and DoF equivalence make sense to a certain extent. How about the rest of photography like exposure?

If you use the same exposure parameters on FF and C2 (Full frame and crop 2) and have the same field of view (ie. 50mm on FF, 25mm on C2):

1. FF DOF will be more shallow
2. FF wil capture four times more light
3. The above two are a tradeoff - to capture more light over fixed unit of time you need to reduce the DOF. There are no ways around this.
4. FF has four times the signal, thus twice the SNR
5. Thus the images of FF and C2 will be different using the same exposure parameters.

In order to create an identical omage on different formats you need to use different exposure parameters.

---------- Post added 18-10-14 at 14:33 ----------

This nonsense is going nowhere. Let's wrap it up with what we have proven beyond reasonable doubt:

Fact 1: "Equivalence" (in digital photography) IS NOT a method of comparison but a matter of faith. It's like religion.

Equivalency as a tool for comparing different formats. Nothing more nothing less. It is based on solid science. You may not like it, but it that doesn't make it into religion.

Here is a small example:

1. FF sensor (full frame) and C2 (crop factor 2) are to be considered
2. They are idealized - identical quantum efficiency, same pixel count, same colour filters, same read noise etc.
3. The above is just to make isse as clear as possible - the real world cameras don't change the results much and the principle at all
4. 25mm lens for the C2, 50mm lens for the FF
5. Same field of view for both due to point 4
6. Exposure time 1 seconds
7. Aperture number of f/2 is considered
8. The FF aperture if thus 25mm, while the C2 aperture is 12.5mm
9. The area of the apertures differ by factor of 4 (four) - area = pi*r^2
10. Four (4) times more light goes through the bigger aperture even though the aperture number is the same
11. Standard deviation (ie. noise) of light is the square root of the number of photons captured (poisson distribution)
12. FF has four times more light, thus four times more signal.
13. FF has two times more noise (!)
14. FF has two times higher signal-to-noise-ratio (SNR)
15. FF also has more shallow DOF - this is the tradeoff that is needed for higher SNR with fixed exposure time and light

1. If we change the FF aperture number to f/4:
2. Same aperture diameter
3. Same amount of light captured
4. Same noise
5. Same DOF
6. Identical image information (ie. apart from the inherit randomness of the light and sensor noise the images are identical)
7. The FF could be exposud for up to four (4) times longer to get better SNR unless for example movement prevents this

---------- Post added 18-10-14 at 14:45 ----------

The 5D pixel pitch is 8 um, the K-5 is 4.8 um. The modern tech at work here in the K-5 is attempting to level the playing field of having only a third of the light gathering area per pixel (and the resulting s/n ratio) to work with. And it's seems to almost balance things out. Almost...

Is that what you're saying? 'Cause that's what's happening. [edit: I think I'm agreeing with your disagreement in your post... bigger != newer]

When I started reading this thread I did a total eye-roll / mouth-barf, but I decided to keep reading for the entertainment and limited science vs myth value... If only Jamie and Adam were here; they'd blow something up...

Pixel size is almost irrelevant.
It's the total light that is captured which is relevant, not the light a single pixel captures. The image is made from all the pixels, not one.

The QE (quantum efficiency - the percent of photons entering the sensor (after the color filters etc.) turned into signal) of K5 is about twice that of 5D which means that in spite of the sensor is about 2.25 times smaller it still captures a similar amount of photons with the same exposure parameters.

The deep shadows of the Pentax will be much regardless and actually have signal long after the Canon hasn't any.

Now, if we're not contrained by the exposure parameters, then K5 can capture images with higher SNR all over the place because it has slightly higher sum of pixel's full well capacity. If we saturate both sensors with ETTR, then the K5 has about 8-9% higher SNR over the upper end and middle parts of the tonal range, while in the shadow range it's SNR is even better and infinitely better for the last three stops or so once the Canon loses signal alltogether (at the base ISO).

[jsherman999]

Assuming that the sensor related noises (which are minute and irrelevant outside of deep shadows) are the same, then he is right. The amount of noise (ie. standad deviation) in light itself is equal to the square root of the number of photons. 10000 photons captured means standard deviation of 100 and SNR of 100.


Didn't you just disagree with that?


While the exposure is maintained, the effect of exposure will be different. An f/2.8 image on an iPhone will have vastly lower signal-to-noise ration than an f/2.8 image on a medium format camera.

If you disagree what I wrote in bold, would you explain why it would be wrong.

If you want to have the same SNR on a smaller sensor, you will either need to increase the aperture (ie. reduce the aperture number), or increase the exposure time. Just like you said: it's abut the total light.

---------- Post added 18-10-14 at 14:12 ----------



If you use the same exposure parameters on FF and C2 (Full frame and crop 2) and have the same field of view (ie. 50mm on FF, 25mm on C2):

1. FF DOF will be more shallow
2. FF wil capture four times more light
3. The above two are a tradeoff - to capture more light over fixed unit of time you need to reduce the DOF. There are no ways around this.
4. FF has four times the signal, thus twice the SNR
5. Thus the images of FF and C2 will be different using the same exposure parameters.

In order to create an identical omage on different formats you need to use different exposure parameters.

---------- Post added 18-10-14 at 14:33 ----------



Equivalency as a tool for comparing different formats. Nothing more nothing less. It is based on solid science. You may not like it, but it that doesn't make it into religion.

Here is a small example:

1. FF sensor (full frame) and C2 (crop factor 2) are to be considered
2. They are idealized - identical quantum efficiency, same pixel count, same colour filters, same read noise etc.
3. The above is just to make isse as clear as possible - the real world cameras don't change the results much and the principle at all
4. 25mm lens for the C2, 50mm lens for the FF
5. Same field of view for both due to point 4
6. Exposure time 1 seconds
7. Aperture number of f/2 is considered
8. The FF aperture if thus 25mm, while the C2 aperture is 12.5mm
9. The area of the apertures differ by factor of 4 (four) - area = pi*r^2
10. Four (4) times more light goes through the bigger aperture even though the aperture number is the same
11. Standard deviation (ie. noise) of light is the square root of the number of photons captured (poisson distribution)
12. FF has four times more light, thus four times more signal.
13. FF has two times more noise (!)
14. FF has two times higher signal-to-noise-ratio (SNR)
15. FF also has more shallow DOF - this is the tradeoff that is needed for higher SNR with fixed exposure time and light

1. If we change the FF aperture number to f/4:
2. Same aperture diameter
3. Same amount of light captured
4. Same noise
5. Same DOF
6. Identical image information (ie. apart from the inherit randomness of the light and sensor noise the images are identical)
7. The FF could be exposud for up to four (4) times longer to get better SNR unless for example movement prevents this

---------- Post added 18-10-14 at 14:45 ----------



Pixel size is almost irrelevant.
It's the total light that is captured which is relevant, not the light a single pixel captures. The image is made from all the pixels, not one.

The QE (quantum efficiency - the percent of photons entering the sensor (after the color filters etc.) turned into signal) of K5 is about twice that of 5D which means that in spite of the sensor is about 2.25 times smaller it still captures a similar amount of photons with the same exposure parameters.

The deep shadows of the Pentax will be much regardless and actually have signal long after the Canon hasn't any.

Now, if we're not contrained by the exposure parameters, then K5 can capture images with higher SNR all over the place because it has slightly higher sum of pixel's full well capacity. If we saturate both sensors with ETTR, then the K5 has about 8-9% higher SNR over the upper end and middle parts of the tonal range, while in the shadow range it's SNR is even better and infinitely better for the last three stops or so once the Canon loses signal alltogether (at the base ISO).

A perfect summary, worth reading again!

[kh1234567890]

A perfect summary, worth reading again!

I wouldn't bother.

[Kunzite]

Equivalence is a flawed tool based on solid science. That, when it is based on solid science - I've seen versions which definitely aren't.

Regarding point 2, it reminds me of a joke:
Milk production at a dairy farm was low, so the farmer wrote to the local university, asking for help from academia. A multidisciplinary team of professors was assembled, headed by a theoretical physicist, and two weeks of intensive on-site investigation took place. The scholars then returned to the university, notebooks crammed with data, where the task of writing the report was left to the team leader. Shortly thereafter the physicist returned to the farm, saying to the farmer, "I have the solution, but it only works in the case of spherical cows in a vacuum".

[jsherman999]

I wouldn't bother.

You never are really required to learn beyond the bare scaffolding of facts needed to bring you A) basic sustenance, B) some measure of material surplus beyond A, and C) internet service.

The modern world provides all your needs, no matter how hostile you are to the underlying structures of it. Shoot happy.

---------- Post added 10-18-14 at 02:28 PM ----------

Equivalence is a flawed tool based on solid science. That, when it is based on solid science - I've seen versions which definitely aren't.

Regarding point 2, it reminds me of a joke:
Milk production at a dairy farm was low, so the farmer wrote to the local university, asking for help from academia. A multidisciplinary team of professors was assembled, headed by a theoretical physicist, and two weeks of intensive on-site investigation took place. The scholars then returned to the university, notebooks crammed with data, where the task of writing the report was left to the team leader. Shortly thereafter the physicist returned to the farm, saying to the farmer, "I have the solution, but it only works in the case of spherical cows in a vacuum".

Here's another one:

Guy walks into a bar. Several physicists, mechanical engineers and auto mechanics are having a beer discussing the various ways they can measure and control the horsepower of their car engines, and how to create new engines with more HP for the same displacement, etc.... Guy bellies up to the bar, orders a wine cooler, and loudly exclaims that the other guys should shut up with all the horsepower talk, because some people might take their discussion as an indication that all that matters is horsepower, that the only measure of a car is horsepower, and because of this possibility, any discussion of it is detrimental to the overall automobile industry and a distraction for real drivers.

The mechanics, engineers and physicists pat the guy on the head, order him another wine cooler and tell him they'll think about it.

Guy is happy, drives home.

(Of course the funny part of that is the end - we all know the guy won't drive home, just go to another bar looking for another HP-offensive discussion to shut down! )

[Kunzite]

Interesting story, but I like mine better - it fits the context. Yours only fits a false vision about "equivalentionits" being somehow the experts of the photographic world...
Unless those "physicists, mechanical engineers and auto mechanics" were discussing not about horse power, but cow power. Spherical cow in a vacuum power.

[Ian Stuart Forsyth]

You never are really required to learn beyond the bare scaffolding of facts needed to bring you A) basic sustenance, B) some measure of material surplus beyond A, and C) internet service.

The modern world provides all your needs, no matter how hostile you are to the underlying structures of it. Shoot happy.

---------- Post added 10-18-14 at 02:28 PM ----------





Here's another one:

Guy walks into a bar. Several physicists, mechanical engineers and auto mechanics are having a beer discussing the various ways they can measure and control the horsepower of their car engines, and how to create new engines with more HP for the same displacement, etc.... Guy bellies up to the bar, orders a wine cooler, and loudly exclaims that the other guys should shut up with all the horsepower talk, because some people might take their discussion as an indication that all that matters is horsepower, that the only measure of a car is horsepower, and because of this possibility, any discussion of it is detrimental to the overall automobile industry and a distraction for real drivers.

The mechanics, engineers and physicists pat the guy on the head, order him another wine cooler and tell him they'll think about it.

Guy is happy, drives home.

(Of course the funny part of that is the end - we all know the guy won't drive home, just go to another bar looking for another HP-offensive discussion to shut down! )

Made me laugh

[ElJamoquio]

Interesting story, but I like mine better - it fits the context. Yours only fits a false vision about "equivalentionits" being somehow the experts of the photographic world...
Unless those "physicists, mechanical engineers and auto mechanics" were discussing not about horse power, but cow power. Spherical cow in a vacuum power.

An engineer is probably not the best driver in the world, but the engineer can tell the driver and calibrator what engine speed, cam position, and air-fuel ratio at which to operate the engine for best performance.

[jsherman999]

An engineer is probably not the best driver in the world, but the engineer can tell the driver and calibrator what engine speed, cam position, and air-fuel ratio at which to operate the engine for best performance.

And it's a very good thing some folks are paying attention to such things...

(Kunzite, I'll buy you that wine cooler.)

[Kunzite]

There's a good reason that the spherical cow joke it's about a theoretical physicist, like with your much beloved "equivalence" he got an answer which can only be applied in an "idealized", theoretical world. You are not the "mechanical engineers", but the "theoretical physicists" of the photographic world.