[beholder3]

Let's try it in simple words with as little scientific language as possible with a simple simulation.

Remember that "dynamic range" in photography is the size of the spread between the brightest bits of information in a photo and the darkest bits.
Dynamic range - Wikipedia

What are the limiting factors at each end?

• bright end:the sensor pixels at some point can not register more photons and are "full". The image area now is pure white and no more detail can be recovered.
• dark: the (mostly random) noise in the really dark areas can grow so much that you are unable to get meaningful image subject information. The nerds call this signal-to-noise-ratio ("SNR") being very low. At SNR 1 (=1:1) it means you have as much good image information ("signal") as you have crappy noise.

So the space between clipping highlights and massive noise in the shadows is dynamic range.

Comparison setup to be realistic:

1. Imagine the exact same exposure parameters to be used for both a large and a small sensor camera.
   For example it could be ISO 100, shutter 1 sec., aperture F/8 (not "equivalent") on both cameras.
2. we ignore DoF for this question for now as it is about maximum dynamic range only
3. Imagine shooting the absolute black subject. We do not need to worry about lenses here.
4. To make the point very visually easy here, let us assume a major size difference between the sensors. Crop factor 8.
5. Imagine both sensor have the exact same technology level and the exact same pixel size. So the individual pixel is technically the same.
   This means the pixels will clip to white at exactly the same brightness level. So the upper end of the dynamic range is the same for both sensors.

Look at what the small sensor has registered:



Hm. You might barely notice some little noise here.

Now let us take a look at what the large sensor has registered:



It is 8 times as wide and 8 times as high as the smaller sensor. To simulate the same technology, same exposure I just created a 8x8 mosaic from the smaller sensor. So both sensors work absolutely identical.

Still looking only very mildly noisy.


If you look at the files in 100% screen mode you will see proof that the noise is exactly the same on both per area (!).

User style option A)

And if you are a person who typically watches images as I have presented the two pieces above - with the display sizes as different as the sensor sizes - then this is it for you. Same dynamic range for both. No need to worry about "equivalency".

User style option B)

But actually this is not how the vast majority of people watch images. They typically look at both images on the same electronic device (screen) or in the same print size.

The latter obviously means you have to enlarge the crop 8 sensor image 8 times more. And this has quite an effect:

Left side: large sensor, right side small sensor




Squint your eyes a little bit and look at both images. You will see the noise on the crop 8 sensor is much worse.

This does show that the magnification path from sensor size to display size actually changes the perceived noise.

Magnifying an image means raising the noise level. This in itself means lowering the dynamic range by magnification.
Because larger sensors do need less magnification from sensor to display sizes the perceived noise is lower and the dynamic range higher with regards to the displayed image - while it stays the same per sensor area.

Yes, even while the small sensor image has the exact same level of noise per sensor area, the magnification from sensor to display makes it worse for the watching person. It's all about magnification.


Enjoy the discussions.

Attached Images

     

[photoptimist]

Yep! Comparing a 1 megapixel image blown up to 8"X10" and a 64 megapixel image downsampled to 8"x10" is not unlike comparing a 110 format negative blown up to a contact print of a 8x10 negative of the same film base. A given film base will have the same intrinsic grain properties regardless of the size of negative, but the final result will look much nicer with a large format negative.

A huge part of the "equivalence" debate is in disagreement about what's apples-to-apples, what's oranges-to-oranges, and what's apples-to-oranges when it comes to different pixel sizes, different sensor sizes, different focal lengths, different viewing conditions, etc. that are unavoidable when comparing images of different formats.

[StiffLegged]

Any more "equivalence" blather deserves pelting with rotten apples, squishy oranges or wet aubergines until they all promise to shut up and go away.

It's not personal, just not interested, over and out.

[swanlefitte]

Makes sense.
There will be an objective floor where the noise isn't perceived and a subjective ceiling where the noise is deemed too much in both.
Thought this might be a point that could pertain if situational points are brought up as this is about the space of the perceived noise.

[Dartmoor Dave]

Yep! Comparing a 1 megapixel image blown up to 8"X10" and a 64 megapixel image downsampled to 8"x10" is not unlike comparing a 110 format negative blown up to a contact print of a 8x10 negative of the same film base. A given film base will have the same intrinsic grain properties regardless of the size of negative, but the final result will look much nicer with a large format negative.

Fair enough. And the difference between, let's say, a 24 megapixel APS-C sensor and a 24 megapixel FF sensor will be negligible for most practical purposes after downsampling.

Because that's where this thread will inevitably head. The equivalentists will try to erect an edifice of such baroque absurdity on this foundation that popcorn sellers will jump for joy as folk gather to watch on in horror.

[biz-engineer]

dark: the (mostly random) noise in the really dark areas can grow so much that you are unable to get meaningful image subject information. The nerds call this signal-to-noise-ratio ("SNR") being very low. At SNR 1 (=1:1) it means you have as much good image information ("signal") as you have crappy noise.

Level Below 1/2 lsb translates into value 0 of ADC output as long as the peak value of noise (shot noise + thermal noise) is lower than 1/2 lsb. Therefore, a 14bits ADC can't code more then 14 ev of light level if photons -> voltage transfer function is a straight line. Even with zero noise, with a 14bits ADC you get 14 ev of dynamic range maximum. You can decimate multiple values of 14bits to make a more precise value, but it only works if there is enough noise in the system, otherwise the ADC value is stuck at 0 and the average of zeros is zero.

[stevebrot]

What are the limiting factors at each end?

Linking dynamic range to S/N ratio is classic equivalence talk, but ignores the elephant in the room, that being that digital capture is intrinsically lacking for tonal rendition of low values. This is true even in the total absence of noise. Low value grays are not only in danger of clipping (no recorded value -- 0), but as values approach zero they do so in abrupt steps such that the bottom three stops of dynamic range (1 - 23) are accomplished with only eight tones, best case. If dynamic range is defined in terms of detail capture (driven by tonal rendering), the actual span is much narrower.


Steve

---

Last edited by stevebrot; 10-27-2019 at 10:26 AM.

[biz-engineer]

Magnifying an image means raising the noise level. This in itself means lowering the dynamic range by magnification.

Really? Does magnification change the difference in magnitude between the darkest point and the brightess point?

[photoptimist]

Fair enough. And the difference between, let's say, a 24 megapixel APS-C sensor and a 24 megapixel FF sensor will be negligible for most practical purposes after downsampling.

Because that's where this thread will inevitably head. The equivalentists will try to erect an edifice of such baroque absurdity on this foundation that popcorn sellers will jump for joy as folk gather to watch on in horror.

There's some truth to what you say.

Never underestimate humanity's willingness to pay 100% more for a 1% improvement that lets them shout "my gear is top-ranked!" The measurebators and "rank" amateurs only want the best-of-the-best. I'd like to say "let them have their fun" if only they weren't so noisy about it.

That said, I really do enjoy the IQ of the K-1. Another stop of high-ISO performance is surely welcome when shooting in poor ambient light.

[Dartmoor Dave]

Let's try it in simple words

Here's a couple of simple words: upsampling and downsampling.

In your examples you've taken a 50x50 pixel image and a 400x400 pixel image, then upsampled the 50x50 pixel version to 400x400. In the case of any recent digital camera, using any display technology currently available, we will always be looking at a downsampled image, rendering your examples irrelevant.

For your examples to be in any way meaningful, photographers would need to be routinely looking at their photos at greater than 100% on their screens. And who the heck ever views photos at 200% or more?

[MatKus]

Probably that's why we have not 64 and even 108 MPX sensors on smartphones, yest you can't make with them image bigger, than ~16MPX. They realy make over 100 megapixels sensors, just they average nearby pixels to make noise harder to notice. Signal is same at every pixel, noise is random - sometimes pixel is too bright, sometimes too dark, average should be around 0 in theory.
And sometimes those sensors can't even be read at their full resolution, they average image in sensor electronics.
I wander if we could do this kind of "cheating" in "big" cameras.

[stevebrot]

Here's a couple of simple words: upsampling and downsampling.

I agree

Upsampling = extrapolated data (mathematical best guess for what was never captured)

Downsampling = judicious pixel removal (mathematical best guess for retaining the gist of what was captured)

For fun and games, one can do a 2X upsample followed by a 0.5X downsample and compare the end result with the original image of the same pixel dimensions. Then do the reverse for complete amusement.

For your examples to be in any way meaningful, photographers would need to be routinely looking at their photos at greater than 100% on their screens. And who the heck ever views photos at 200% or more?

Several members on this site do image evaluation on a 2X upsample, perhaps due to use of high pixel pitch monitors where anything not enlarged is sooooo very teeny! I have long felt that photo editing should be done with monitor resolution on the verge of the individual pixels being visible to the user.


Steve

[UncleVanya]

Any more "equivalence" blather deserves pelting with rotten apples, squishy oranges or wet aubergines until they all promise to shut up and go away.

It's not personal, just not interested, over and out.

Why post and not simply hit the ignore thread button? Some people may find this discussion relevant and/or entertaining.

[swanlefitte]

I guess we have to post something like this on every geek topic. This is Rodger from Lensrentals header on a recent blog post.


"GEEK ALERT!!

Let’s be absolutely clear; this is not a practical or useful article. It won’t help your photography or cinematography become better. It won’t help you choose equipment any time in the next couple of years. It won’t provide any fodder for your next Forum War. It’s just a geek article that may interest some people. It may give a little peek into what may come in the future, and some insight into the kind of work we’re actually doing behind the scenes at Olaf. So if you’re interested in that kind of stuff, read along. "

[StiffLegged]

Why post and not simply hit the ignore thread button? Some people may find this discussion relevant and/or entertaining.

I wasn’t aware of the Ignore Thread option but will avail myself of it. Thanks for the tip! :-)