Shot noise question

[Shashin]

Good question, but I forgot to mention that I was assuming equal viewing distance of all the outputs mentioned in the post you (partly) quoted.

This for me is a very important question. If you view a print from two feet away and then three feet away, is the noise in the image perceptually different? That would be the same as as looking an APS-C and 35mm image at one viewing distance. I suspect, only because I view my prints from different distances, there is no perceptual difference.

 

[MGrayson]

Will,

What is this Equivalency Principle? There’s the theorem about scaling sensor size, focal length, F stop, and ISO. But that’s theoretical, not practical. Or do they mean something else?

Matt

 

[Shashin]

However, at a constant print size, you'll have more pixels contributing to each square mm on the print with the larger sensor, and thus the SNR will be higher for the larger sensor.

I would say you are conflating a signal to noise measurement with the perception of an image: a regression to the mean? This was the issues with measuring granularity in film: granularity was a perceptual measure, not a measure for grain size. SNR is fixed with the system, the perception of that results from a different process (in your example, the SNR could be the same at different viewing distances and assuming equal viewing distance might have statistical validity in one population of photographers, but not in all). This problem is very much like difference in the absolute perspective of an image being the result of camera to subject distance, and the apparent perspective in an image being related to the viewing distance of a print/display. One of these is the result of the imaging system, the other the viewer. But it would be incorrect to say that apparent perspective is the result of focal length, even though focal length and viewing distance correlates.

 

[Shashin]

Will,

What is this Equivalency Principle? There’s the theorem about scaling sensor size, focal length, F stop, and ISO. But that’s theoretical, not practical. Or do they mean something else?

Matt

That is the one.

 

[dougpeterson]

There are many things on your list which don't affect dynamic range measured at the raw file, but I take your meaning.

Absolutely. I should have clarified that the list was a reasonably inclusive list of components, not a list of components specific to dynamic range.

Though some matter that you might not expect. For example the Phase One IQ3 100mp Trichromatic has better shadow quality (in a "what can I print and it looks good/natural?" sense) than the IQ3 100mp despite having every item in the chain identical except the CFA. The tighter CFA led to better color accuracy of shadows post demosaic.

To get an idea of what kinds of differences can occur with the same sensor in four different cameras, take a look at this:

http://photonstophotos.net/Charts/P...elblad H6D-50c,Hasselblad X1D-50c,Pentax 645Z

Jim

Thanks for the link! Very interesting. I'd be even more interested in a semi-qualitative analysis that shows what the resulting shadows look like. But such testing is far harder to come by and far easier to mess up.

 

[JimKasson]

More pixels, more pixels with noise in the absolute sense, but why would the ratio be any higher?

Imagine a 3000x4000 px sensor with a FWC of 100000 e- exposed at a level that produces 16000 electrons. SNR is sqrt(16000) = 126. Now imagine a 6000x8000 px sensor with the same FWC exposed at the same level. Single pixel SNR is still 126. But say we print the images on a 3000x4000 pixel printer. The per pixel SNR of the 3Kx4K camera is 126. But the per pixel SNR of the larger sensor is sqrt(64000) = 253, because four pixels of the larger sensor are combined to make one pixel of the print.

Jim

 

[Abstraction]

Would it be correct to say that shot noise is a theoretical SNR limit and that for all practical purposes, with all else being equal, it's the pixel size that has more of a bearing on the SNR than shot noise? A bigger sensor at a given resolution has bigger pixels, thereby having a greater SNR ratio than a smaller sensor and a bigger sensor given the same pixel pitch will have greater resolution, therefore, a print can be larger at a given SNR than a smaller sensor with the same pixel pitch.

Do I have this down?

 

[JimKasson]

I would say you are conflating a signal to noise measurement with the perception of an image: a regression to the mean? This was the issues with measuring granularity in film: granularity was a perceptual measure, not a measure for grain size. SNR is fixed with the system, the perception of that results from a different process (in your example, the SNR could be the same at different viewing distances and assuming equal viewing distance might have statistical validity in one population of photographers, but not in all). This problem is very much like difference in the absolute perspective of an image being the result of camera to subject distance, and the apparent perspective in an image being related to the viewing distance of a print/display. One of these is the result of the imaging system, the other the viewer. But it would be incorrect to say that apparent perspective is the result of focal length, even though focal length and viewing distance correlates.

SNR of the file sent to the printer is not fixed with the system. The way it changes from the SNR on the the sensor depends on the relationship of the resolution of the sensor to that of the file sent to the printer. This is unrelated to psychology.

If we get into perception, the CSF of the viewer and the viewing distance both become important:

https://blog.kasson.com/the-last-word/contrast-sensitivity-functions-and-photography/

Jim

 

[JimKasson]

Would it be correct to say that shot noise is a theoretical SNR limit and that for all practical purposes, with all else being equal, it's the pixel size that has more of a bearing on the SNR than shot noise? A bigger sensor at a given resolution has bigger pixels, thereby having a greater SNR ratio than a smaller sensor and a bigger sensor given the same pixel pitch will have greater resolution, therefore, a print can be larger at a given SNR than a smaller sensor with the same pixel pitch.

Do I have this down?

Pretty much.

Is is correct to say that the shot noise determines the theoretical SNR limit. The shot noise is not the limit itself, just the denominator of the limit. PRNU, for example, can degrade the SNR from the Poisson ideal for large signals, and read noise for the small ones.

Bigger pixels don't necessarily mean higher shot noise limited SNR, but usually, bigger pixels means higher full well capacities, and that does mean higher shot noise limited SNR.

Comparison at different print sizes is a bit complex, and would need a longer discussion.

Jim

 

[JimKasson]

This for me is a very important question. If you view a print from two feet away and then three feet away, is the noise in the image perceptually different?

It is, because of the human luminance and choma contrast sensitivity functions. They filter noise spatially.

Jim