Does dynamic range (DR) affect noise? (Page 6): -----------------------------(reply) Boo---------------now get your last word on that word?----------------------...

OldSchool-WI Loc: Brandon, Wisconsin 53919

SuperflyTNT wrote:

I read what you wrote. I don’t need to build cameras and sensors to know how they work.

-----------------------------(reply)

Boo---------------now get your last word on that word?----------------------

bclaff Loc: Sherborn, MA (18mi SW of Boston)

Strodav wrote:

...I started thinking of the makeup of the Bayer filter with 2x green samples compared to red and blue and that a relatively sophisticated algorithm is used to convert RGGB, and maybe some other surrounding samples, into a single output pixel. They are at least using 14 bits of red, 14 bits of blue and two 14 bits samples of green. So I'm thinking a bit more than 2^14 DR ...

The demosaicing process does consult neighboring pixels in a sophisticated way, but this doesn't change the underlying dynamic range.

Consider ultra simple demosaicing that combines the 4 pixels of a typical Color Filter Array (CFA).
Mathematically you would have 4x the signal (saturation) and 2x the noise (noise floor) so you might naively think you've got 2x (4x / 2x) more dynamic range. But you also increased the area of your "pixel" 4x! The combination has no effect on the dynamic range per unit area. The dynamic range of the underlying sensor pixel is not increased.

This is also the reason the pixel size has no effect on something like PhotonsToPhotos Photographic Dynamic Range or DxOMark Dynamic Range score ("print" dynamic range).

selmslie Loc: Fernandina Beach, FL, USA

OldSchool-WI wrote:

Totally absent in this technical discussion is that film emulsions were more than the straight lines within the exposure range of digital. There were characteristic curves which played a major part in the quality of the rendition. And those were quite known and public for every emulsion.---------------

I did not mention film in this thread but I described its behavior a couple of days earlier on page 3 of The "Fear of Noise".

That's where I mentioned that noise is more visible in the digital mid-tones for the same reasons that we can see grain more clearly in film mid-tones - the "S" curve. It's the higher contrast in the final image's mid-tones that makes that happen.

selmslie Loc: Fernandina Beach, FL, USA

TriX wrote:

It seems to me this is a plot of the linearity of your camera, which is quite good until you get down to where that old bugaboo, noise (and perhaps non linearities or quantization errors of the A/D), intrudes

I will run the same test with the Z7 at its lower base ISO 64 since it also offers both a 14- and 12-bit raw capture.

I no longer have the D610 but I will see if I can find a 12- to 14-bit comparison when I checked its linearity.

My A7 II no longer has its Bayer array and it only captures 14-bit raw but I will also run the test to see where the linearity starts to fail.

bclaff Loc: Sherborn, MA (18mi SW of Boston)

selmslie wrote:

... I have been thinking of a different manifestation of DR that is independent of noise. ...
if you examine the numbers in the attached spreadsheet you will see that after Ev-5 the both sets of raw values get a bit squirrely.

That's due to noise, so you have not in fact made a test that is "independent of noise".

selmslie wrote:

... By that crude measure the Df seems to have a reliable Df of about 9 stops for the 14-bit raw file but only about 7 stops for the 12-bit raw file.

If I understand your protocol you're implicitly at middle gray (18%) so you need to add about 2.5 stops to your estimate. 9 + 2.5 = 11.5 which is getting in the neighborhood of the 12.9 stops of EDR I measure at 14-bits for the Nikon Df. And yes, 12-bits would produce a lower value probably 12 or 12.5 at best.

selmslie Loc: Fernandina Beach, FL, USA

bclaff wrote:

The demosaicing process does consult neighboring pixels in a sophisticated way, but this doesn't change the underlying dynamic range.

I should have mentioned that the raw values I have been recording were taken from RawDigger's default spot 150x100 pixels selection at the center of the frame. That's an average of the values of 15,000 pixels and noise becomes obvious when the standard deviation gets close to the average value.

To save time I only took the values from the G1 channel until the change in log(2) of the raw values started to deviate by more than about 1 stop. Once that happened I took both the G1 and G2 values and averaged them as you can see in the spreadsheet.

bclaff Loc: Sherborn, MA (18mi SW of Boston)

selmslie wrote:

I will run the same test with the Z7 at its lower base ISO 64 since it also offers both a 14- and 12-bit raw capture.

I no longer have the D610 but I will see if I can find a 12- to 14-bit comparison when I checked its linearity.

My A7 II no longer has its Bayer array and it only captures 14-bit raw but I will also run the test to see where the linearity starts to fail.

Test by all means but you are going to find that all digital camera pixels are quite linear with respect to how much light is collected. This is really a basic premise of the technology. Pixel response is non-linear as the pixel saturates but often the gain before the Analog to Digital Converter (ADC) is set in such as way to only use the linear portion of the pixel response. At the low end read noise will interfere with your observing the linearity of the pixel. To see why you might want to look at some Photons Transfer Curves (PTCs) at PhotonsToPhotos.

selmslie Loc: Fernandina Beach, FL, USA

bclaff wrote:

If I understand your protocol you're implicitly at middle gray (18%) so you need to add about 2.5 stops to your estimate. 9 + 2.5 = 11.5 which is getting in the neighborhood of the 12.9 stops of EDR I measure at 14-bits for the Nikon Df. And yes, 12-bits would produce a lower value probably 12 or 12.5 at best.

When I was testing linearity I noticed that some of my cameras (like the Df) could be exposed at up to 4 stops above the camera's middle gray without reaching the raw limit. Others could not go that high. I'll have to look through my notes to see how close the D610 and A7 II got (when it still had its Bayer array).

The linearity testing I did back then was looking for the level where highlight warnings started in each camera. The Nikons behaved similarly (at about 0.7 stops before the upper limit) but the Sony was more conservative at about 1.3 stops.

bclaff Loc: Sherborn, MA (18mi SW of Boston)

selmslie wrote:

I should have mentioned that the raw values I have been recording were taken from RawDigger's default spot 150x100 pixels selection at the center of the frame.

FWIW, my RawDigger doesn't have a default selection. Perhaps yours is persisted from an earlier session. But you can always use Selection/set Selection by Numbers

selmslie wrote:

... noise becomes obvious when the standard deviation gets close to the average value.

You're implicitly referring to the Signal to Noise Ration (SNR). And yes, the lower the SNR the more apparent noise. PhotonsToPhotos Photographic Dynamic Range uses a minimum SNR in addition to the Circle of Confusion. That's an important point I omitted earlier.(COC)

selmslie wrote:

To save time I only took the values from the G1 channel until the change in log(2) of the raw values started to deviate by more than about 1 stop. Once that happened I took both the G1 and G2 values and averaged them as you can see in the spreadsheet.

It's of no practical consequence but you ought to be consistent in your methodology. Remember that when you combine noise it should be sqrt((g1^2 + g2^2) / 2) not (g1 + q2) / 2

bclaff Loc: Sherborn, MA (18mi SW of Boston)

selmslie wrote:

When I was testing linearity I noticed that some o... (show quote)

I suggest that you take metering entirely out of the equation and simply use exposure time to vary the amount of light collected. Depending on the camera there are caveats.

selmslie Loc: Fernandina Beach, FL, USA

bclaff wrote:

Test by all means but you are going to find that all digital camera pixels are quite linear with respect to how much light is collected. ...

It's not just the pixels. Another consideration is the shutter accuracy and precision.

The shutter on the Df is very good at long exposures but, like all cameras, the precision and accuracy at very high shutter speeds may be contributing to a lack of linearity.

I added a crude test for noise based on the LOG(2) ratio of average raw value within the 15k selection to the standard deviation shown in RawDigger.

This may help explain why the noise looks pretty much the same for a 12- and 14-bit capture of a scene with a DR of about 9 stops which is all that most of us really care about. I will follow the same process when I get to the other two cameras.

Here us the updated spreadsheet:

Raw values for 12- and 14-bit capture over 14 stops.xlsx (revised)
Attached file:
(Download)

selmslie Loc: Fernandina Beach, FL, USA

bclaff wrote:

FWIW, my RawDigger doesn't have a default selection. Perhaps yours is persisted from an earlier session. But you can always use Selection/set Selection by Numbers

That's what I am referring to. It defaults to a 150x100 selection in the upper left corner and the checkbox centers it.

I used the camera's spot meter, also at the center, to get the initial Ev+0 shutter speed.

The spot meter does not cover the same area as RawDigger's selection but at least they are both centered.

bclaff wrote:

It's of no practical consequence but you ought to be consistent in your methodology. Remember that when you combine noise it should be sqrt((g1^2 + g2^2) / 2) not (g1 + q2) / 2

I wasn't combining noise. I was just getting the raw values. At about 5 stops below middle gray G1 and G2 start to disagree (in the Df).

Ysarex Loc: St. Louis

OldSchool-WI wrote:

____________________________(reply)
Jealousy jealousy. Maybe some day you will advance from consumer level to pro-level Sigma equipment? In the meantime, don't show your ignorance--it is not becoming and oozes out with each of your posts.---------------------Happy new year.

No photo? Of course not -- just worthless talk & no show.

selmslie Loc: Fernandina Beach, FL, USA

bclaff wrote:

I suggest that you take metering entirely out of the equation and simply use exposure time to vary the amount of light collected. Depending on the camera there are caveats.

Metering is only used to locate the middle gray that the camera is indicating. All of the exposures are shot at full manual by changing only the shutter speed.

Here are the results for the Z7:

An additional stop of exposure would have blown the highlights so I only show 13 data points. The linearity is very consistent over the range we would care about but 14-but still has a couple more useable shadow stops.

S/N ratios are very consistent over the first 7 stops. The 12-bit raw values drop below 1 at Ev-8 so I stopped plotting S/N.

Since the A7 II has no Bayer array the relationship between the meter and middle gray is questionable. There is no 12-bit option. I normally ignore the meter and rely only on the Zebra warnings. I am including these plots because they still tell us something about linearity and S/N ratios.

Because the raw value is down to 1 the data is limited to 12 stops. It is very linear.

The drop in S/N is also linear for about 8 stops before it gets jagged.

The spreadsheet contains all three tests:

Raw values for 12- and 14-bit capture over 14 stops.xlsx
Attached file:
(Download)

bclaff Loc: Sherborn, MA (18mi SW of Boston)

selmslie wrote:

...I was just getting the raw values....

There's no need for the raw values since RawDigger computes the mean and standard deviation of the selection for you. Of course you need to subtract BlackLevel from the mean assuming you have Subtract Black turned off. This is an ILCE-1 30sec black frame.