Pages: 1 [2] 3   Go Down

Author Topic: 42MP x 12-bit dynamic range vs. 20MP x 14-bit dynamic range  (Read 16066 times)

Jack Hogan

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 798
    • Hikes -more than strolls- with my dog
Re: 42MP x 12-bit dynamic range vs. 20MP x 14-bit dynamic range
« Reply #20 on: November 09, 2015, 01:48:04 pm »

The dynamic range of a ADC (or DAC) is the ratio bentween the largest expressible value vs. the smallest expressible value. It's is that simple.

Hello EinstStein,

All the definitions of DR I have seen with respect to imaging sensors refer to some variation of the ratio of maximum usable signal to minimum usable signal.  Often the minimum usable signal is defined according to an application specific SNR, for instance the signal at which SNR = 1.  DxOmark.com uses this latter definition.  In some circles the minimum usable signal is considered to be equal to the read noise (i.e. the noise 'floor' introduced by the electronics).  Some ADC designers use this rule to determine bit depth.  On the other hand I have never seen a definition of DR based on 'expressible value'.   Do you have a source for it?  More information on DR for photographers here.

Jack
Logged

Dave Ellis

  • Full Member
  • ***
  • Offline Offline
  • Posts: 103
Re: 42MP x 12-bit dynamic range vs. 20MP x 14-bit dynamic range
« Reply #21 on: November 10, 2015, 12:05:35 am »

This thread has got me thinking about dynamic range, bit depth and noise and I decided to do a simple test with my D610 as follows.

I set up a fairly high dynamic range shot as shown in the first attachment and took the following variants as shown in the 100% crops.

1. ISO 100, 14 bit  (DXOMark quotes dynamic range for ISO100 of 14.36 stops)

2. ISO 100 12 bit

3. ISO 800 14 bit (DXOMark quotes dynamic range for ISO 800 of 12.23 stops)

All crops were processed with the same settings in ACR ie Exposure +1, Highights -50, Shadows +100

These images indicate that there is very little difference in shadow noise between the 14 bit and 12 bit versions of the ISO 100 shots (14.36 EV).

The ISO 800 shot indicates considerably more noise than the ISO 100 shots corresponding to the lower dynamic range of 12.23 EV. This suggests to me that from a noise point  of view, the 12 bit version of the ISO 100 shot does not show a reduction in dynamic range to 12 EV. Maybe I'm not looking at the whole picture though ?

Please note that the tests were not meant to investigate other issues such as posterisation.

Dave
« Last Edit: November 10, 2015, 12:12:26 am by Dave Ellis »
Logged

ErikKaffehr

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 11311
    • Echophoto
Re: 42MP x 12-bit dynamic range vs. 20MP x 14-bit dynamic range
« Reply #22 on: November 10, 2015, 12:45:02 am »

Hi Dave,

Thanks for making this experiment. Your findings are interesting, but I am quite a bit from drawing any conclusion yet.

Best regards
Erik


This thread has got me thinking about dynamic range, bit depth and noise and I decided to do a simple test with my D610 as follows.

I set up a fairly high dynamic range shot as shown in the first attachment and took the following variants as shown in the 100% crops.

1. ISO 100, 14 bit  (DXOMark quotes dynamic range for ISO100 of 14.36 stops)

2. ISO 100 12 bit

3. ISO 800 14 bit (DXOMark quotes dynamic range for ISO 800 of 12.23 stops)

All crops were processed with the same settings in ACR ie Exposure +1, Highights -50, Shadows +100

These images indicate that there is very little difference in shadow noise between the 14 bit and 12 bit versions of the ISO 100 shots (14.36 EV).

The ISO 800 shot indicates considerably more noise than the ISO 100 shots corresponding to the lower dynamic range of 12.23 EV. This suggests to me that from a noise point  of view, the 12 bit version of the ISO 100 shot does not show a reduction in dynamic range to 12 EV. Maybe I'm not looking at the whole picture though ?

Please note that the tests were not meant to investigate other issues such as posterisation.

Dave
Logged
Erik Kaffehr
 

Dave Ellis

  • Full Member
  • ***
  • Offline Offline
  • Posts: 103
Re: 42MP x 12-bit dynamic range vs. 20MP x 14-bit dynamic range
« Reply #23 on: November 10, 2015, 12:59:44 am »

Hi Dave,

Thanks for making this experiment. Your findings are interesting, but I am quite a bit from drawing any conclusion yet.

Best regards
Erik

Look forward to hearing your thoughts Erik, and others too.

I am assuming that changing from 14 bit capture to 12 bit capture actually changes the bit depth of the adc. Even if this is not the case and the 12 bit is just down sampled from the 14 bit, similar considerations would apply I think.

Dave

Logged

ErikKaffehr

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 11311
    • Echophoto
Re: 42MP x 12-bit dynamic range vs. 20MP x 14-bit dynamic range
« Reply #24 on: November 10, 2015, 01:27:48 am »

Hi Dave,

One consideration may be that your darks may be dominated by shot noise and not by readout noise and one reason for that may be lens flare.

Some knowledgeable authors say that lens flare limits practical DR to 11EV, I would guess it holds but I think I have demonstrated some cases when that assumption is not valid.

A different discussion, but I have not been able to observe any difference between Sony's compressed raw and uncompressed raw, except color shift in the darks in the raw compressed image. Noise levels were similar.

But, Jim Kasson did observe difference between 14-bit and 12-but samples on the Sony A7rII, but I think he was looking at much higher noise levels than you did. I'll try to dig up that article.

Best regards
Erik


Look forward to hearing your thoughts Erik, and others too.

I am assuming that changing from 14 bit capture to 12 bit capture actually changes the bit depth of the adc. Even if this is not the case and the 12 bit is just down sampled from the 14 bit, similar considerations would apply I think.

Dave
Logged
Erik Kaffehr
 

Dave Ellis

  • Full Member
  • ***
  • Offline Offline
  • Posts: 103
Re: 42MP x 12-bit dynamic range vs. 20MP x 14-bit dynamic range
« Reply #25 on: November 10, 2015, 01:45:01 am »

Hi Dave,

One consideration may be that your darks may be dominated by shot noise and not by readout noise and one reason for that may be lens flare.

Best regards
Erik

Erik I had a look at the 14 bit ISO100 file in RawDigger. The darkest areas (between the wooden slats on the chair) have values around the 20-30 mark. I'm not sure I could calculate shot noise with confidence !!

Dave
Logged

bjanes

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 3387
Re: 42MP x 12-bit dynamic range vs. 20MP x 14-bit dynamic range
« Reply #26 on: November 10, 2015, 09:23:43 am »

Erik I had a look at the 14 bit ISO100 file in RawDigger. The darkest areas (between the wooden slats on the chair) have values around the 20-30 mark. I'm not sure I could calculate shot noise with confidence !!

Dave

Assuming saturation at 16383 (may be less in practice), a raw value of 20 is 9.7 stops down from saturation. This is in the range that can be quantified at 12 bits. To see the advantage of 14 bits, you would have to have darker blacks. Correct?

Bill
« Last Edit: November 11, 2015, 06:37:50 am by bjanes »
Logged

AlterEgo

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 1995
Re: 42MP x 12-bit dynamic range vs. 20MP x 14-bit dynamic range
« Reply #27 on: November 10, 2015, 10:13:26 am »

But, Jim Kasson did observe difference between 14-bit and 12-but samples on the Sony A7rII
wasn't that actually between 13 effective bits (with compressed raws, not with the recent totally uncompressed raws) vs 12 bit (where Sony drops to 12 bit in certain modes) ?
Logged

bjanes

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 3387
Re: 42MP x 12-bit dynamic range vs. 20MP x 14-bit dynamic range
« Reply #28 on: November 10, 2015, 10:36:40 am »

Some knowledgeable authors say that lens flare limits practical DR to 11EV, I would guess it holds but I think I have demonstrated some cases when that assumption is not valid.

Hi Erik,

Veiling glare does limit DR. The extent of the limitation depends on the relative proportions of light and dark in the image. Here is a demonstration of an old test I did with the Nikon D3 using a Stouffer wedge. I took images with the wedge with the background masked off (on the right) and unmasked (left).


I think I rendered to TIFFs using ACR with a linear tone curve (PV2010 with sliders on main tab zeroed) and here are the results.


And the DRs determined by Imatest.


With the background masked off, I got 10.6 stops of DR. It would be interesting to repeat the test with the D800e or the A7Rii.

Bill
Logged

Dave Ellis

  • Full Member
  • ***
  • Offline Offline
  • Posts: 103
Re: 42MP x 12-bit dynamic range vs. 20MP x 14-bit dynamic range
« Reply #29 on: November 10, 2015, 01:52:13 pm »

Assuming saturation at 16383 (may be less practice), a raw value of 20 is 9.7 stops down from saturation. This is in the range that can be quantified at 12 bits. To see the advantage of 14 bits, you would have to have darker blacks. Correct?

Bill

Yes I think so Bill. I could repeat the test to try and get lower blacks but given the comments from you and Erik on veiling flare, that may be a waste of time. The tests on veiling flare you provide are interesting and yes it would be nice to see similar results for a D800.

Dave
Logged

ErikKaffehr

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 11311
    • Echophoto
Re: 42MP x 12-bit dynamic range vs. 20MP x 14-bit dynamic range
« Reply #30 on: November 10, 2015, 02:16:56 pm »

Hi,

Yes, I would not argue with that. But, the thing was that he has observed tangible benefits…

Best regards
Erik

wasn't that actually between 13 effective bits (with compressed raws, not with the recent totally uncompressed raws) vs 12 bit (where Sony drops to 12 bit in certain modes) ?
Logged
Erik Kaffehr
 

AlterEgo

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 1995
Re: 42MP x 12-bit dynamic range vs. 20MP x 14-bit dynamic range
« Reply #31 on: November 10, 2015, 02:39:14 pm »

Hi,

Yes, I would not argue with that. But, the thing was that he has observed tangible benefits…

Best regards
Erik

but what if tangible between 13 and 12 and tangible between 14 and 13 will yield a little more tangible between 14 and 12 (not talking about compression artefacts naturally) ? I think I saw somebody saying somewhere that totally uncompressed full 14bit also have some better (as in precise) black levels for RGGB channels in raw or something along those lines ...
Logged

ErikKaffehr

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 11311
    • Echophoto
Re: 42MP x 12-bit dynamic range vs. 20MP x 14-bit dynamic range
« Reply #32 on: November 10, 2015, 11:57:19 pm »

Hi Dave,

The best way to study the effects of 12-bit and 14-bit anologue to digital conversion is to check out the darks. Just shoot something like a bookshelf and underexpose 4-5 stops and pull the dark details by correcting exposure in post processing by 4-5 stops. That eliminates the flare factor.

Best regards
Erik


Yes I think so Bill. I could repeat the test to try and get lower blacks but given the comments from you and Erik on veiling flare, that may be a waste of time. The tests on veiling flare you provide are interesting and yes it would be nice to see similar results for a D800.

Dave
Logged
Erik Kaffehr
 

Dave Ellis

  • Full Member
  • ***
  • Offline Offline
  • Posts: 103
Re: 42MP x 12-bit dynamic range vs. 20MP x 14-bit dynamic range
« Reply #33 on: November 11, 2015, 12:28:09 am »

Hi Dave,

The best way to study the effects of 12-bit and 14-bit anologue to digital conversion is to check out the darks. Just shoot something like a bookshelf and underexpose 4-5 stops and pull the dark details by correcting exposure in post processing by 4-5 stops. That eliminates the flare factor.

Best regards
Erik

Thanks Eric

I actually just did some more test shots this time with the lens cap on and a fast shutter speed to try and just record read noise. I'm currently analysing those with RawDigger but it might also be worth doing something like you suggest to get a better visual of the noise. I'll provide an update when I can.

Dave
Logged

Dave Ellis

  • Full Member
  • ***
  • Offline Offline
  • Posts: 103
Re: 42MP x 12-bit dynamic range vs. 20MP x 14-bit dynamic range
« Reply #34 on: November 11, 2015, 08:52:37 pm »

Thanks Eric

I actually just did some more test shots this time with the lens cap on and a fast shutter speed to try and just record read noise. I'm currently analysing those with RawDigger but it might also be worth doing something like you suggest to get a better visual of the noise. I'll provide an update when I can.

Dave

Here are the results of my latest tests. All three shots were taken with the lens cap on in a dark room with a fast shutter speed. The details of the three shots are

1. ISO 100, 14 bit capture

2. ISO 100, 12 bit capture

3. ISO 800, 14 bit capture

All shots have had exposure increased by 7 stops in pp as well as 100% shadow increase, purely so the noise is visible.

I did an analysis of the sample distributions from RawDigger to estimate Standard Deviation of the noise and hence Dynamic Range. Unlike Canon cameras, Nikon have no offset voltage applied to the black level. This means that negative values of noise voltage are all read as 0. I tried to adjust for this by assuming that the distribution of negative values was the same as the positive values and substituted some of the 0 values with these negative values.

The noise calculation for shot 1 (ISO 100) gave an estimated DR of just over 14 stops, which correlates well with the DXOMark results for this camera. The calcs for shot 3 (ISO 800) also gave good correlation around the 12 stop figure.

With shot 2 (ISO 100 with 12 bit), the only sample values were 0 and 1. This is because of the reduced bit depth which can't resolve this low noise figure reliably. This also shows up in the appearance of the shot.

However given the appearance of the images, these results indicate to me that changing from 14 bit to 12 bit for a camera with DR of around 14 doesn't really change the level of noise to any great extent (and hence there is little change to the DR based on the definition of DR= Sat level/Read Noise level.

I'd welcome further comments please.

Dave
« Last Edit: November 11, 2015, 08:58:42 pm by Dave Ellis »
Logged

bjanes

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 3387
Re: 42MP x 12-bit dynamic range vs. 20MP x 14-bit dynamic range
« Reply #35 on: November 12, 2015, 09:56:25 am »

Hi Dave,

The best way to study the effects of 12-bit and 14-bit anologue to digital conversion is to check out the darks. Just shoot something like a bookshelf and underexpose 4-5 stops and pull the dark details by correcting exposure in post processing by 4-5 stops. That eliminates the flare factor.

Best regards
Erik

Erik,

The underexposed bookshelf method is a good method and easy to perform, and it allows to visualize the noise qualitatively. However, at times quantitive measurements are desired and the wedge with Imatest can do this.

However, does the underexposure eliminate the effect of flare? If the veiling glare of a lens is 0.5% (see Norman Koren), the shadows are contaminated by 0.5% of the highlights regardless of exposure. When you bring up the shadows in the raw converter, the effect of the veiling glare is also magnified to the same extent as would occur with an exposure at a higher luminance. I'm not sure that this is the case, and what do you think?

Regards,

Bill
Logged

Bart_van_der_Wolf

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 8913
Re: 42MP x 12-bit dynamic range vs. 20MP x 14-bit dynamic range
« Reply #36 on: November 12, 2015, 11:37:48 am »

Erik,

The underexposed bookshelf method is a good method and easy to perform, and it allows to visualize the noise qualitatively. However, at times quantitive measurements are desired and the wedge with Imatest can do this.

However, does the underexposure eliminate the effect of flare? If the veiling glare of a lens is 0.5% (see Norman Koren), the shadows are contaminated by 0.5% of the highlights regardless of exposure. When you bring up the shadows in the raw converter, the effect of the veiling glare is also magnified to the same extent as would occur with an exposure at a higher luminance. I'm not sure that this is the case, and what do you think?

Hi Bill,

The veiling glare remains proportional (as a percentage of stray 'bright exposure' added) to the actual level of exposure (as a stepwedge example would show). So underexposing will only show less absolute glare (and shadow detail), but the same amount of relative glare. This of course assumes mo Raw processing prior to writing the Raw datafile.

The conceptual difficulty may be in the 'bright exposure' part, because as exposure becomes less, then the absolute amount of veiling glare also becomes less, unless it stems from a very bright lightsource that remains exceedingly bright/clipped. But otherwise, absolute veiling glare amount is reduced proportionally with exposure amount, as is the actual exposure amount, so their proportions stay the same.

The absolute/fixed amount of veiling glare just weighs heavier when it's added to little signal, that's why we lose Dynamic Range.

Cheers,
Bart
Logged
== If you do what you did, you'll get what you got. ==

bjanes

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 3387
Re: 42MP x 12-bit dynamic range vs. 20MP x 14-bit dynamic range
« Reply #37 on: November 12, 2015, 02:11:21 pm »


I did an analysis of the sample distributions from RawDigger to estimate Standard Deviation of the noise and hence Dynamic Range. Unlike Canon cameras, Nikon have no offset voltage applied to the black level. This means that negative values of noise voltage are all read as 0. I tried to adjust for this by assuming that the distribution of negative values was the same as the positive values and substituted some of the 0 values with these negative values.

The noise calculation for shot 1 (ISO 100) gave an estimated DR of just over 14 stops, which correlates well with the DXOMark results for this camera. The calcs for shot 3 (ISO 800) also gave good correlation around the 12 stop figure.

I'd welcome further comments please.

Dave,

A better way to determine the read noise on Nikons is to use the optical black area of the raw file as Bill Claff describes.

Since you use Rawdigger, you can use it to gain access to the optical black area. In preferences check the masked pixels box and uncheck the 2x2 box. Then open the raw file and zoom to 1000% at the right edge of the image to show the masked pixels as shown here. The masked pixels are the 14 darker columns on the right.



You can select them by the numbers as shown.


You can then read the standard deviations showing the read noise, which is about 1.5 ADU for the green channels.


Hope this helps.

Bill
Logged

ErikKaffehr

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 11311
    • Echophoto
Re: 42MP x 12-bit dynamic range vs. 20MP x 14-bit dynamic range
« Reply #38 on: November 12, 2015, 03:26:14 pm »

Hi Bill,

Eliminates was the wrong word, reduces would be better. The idea was mostly that shooting a low contrast target and underexpose would show up possible differences in rendition.

Regarding the flare factor, if we have a bright part of the image, the size of that bright part plays a big role.

Shooting a Stouffer wedge gives good analytic info, but it is not very easy to shoot it well.

I have also noticed that real photographers often don't like things like numbers, graphs, things measurable, facts and so on. The best way to illustrate anything seems to be an 800x600 JPEG…

Best regards
Erik

Ps. Sorry for the rant…

Erik,

The underexposed bookshelf method is a good method and easy to perform, and it allows to visualize the noise qualitatively. However, at times quantitive measurements are desired and the wedge with Imatest can do this.

However, does the underexposure eliminate the effect of flare? If the veiling glare of a lens is 0.5% (see Norman Koren), the shadows are contaminated by 0.5% of the highlights regardless of exposure. When you bring up the shadows in the raw converter, the effect of the veiling glare is also magnified to the same extent as would occur with an exposure at a higher luminance. I'm not sure that this is the case, and what do you think?

Regards,

Bill
Logged
Erik Kaffehr
 

Dave Ellis

  • Full Member
  • ***
  • Offline Offline
  • Posts: 103
Re: 42MP x 12-bit dynamic range vs. 20MP x 14-bit dynamic range
« Reply #39 on: November 12, 2015, 04:00:40 pm »

Dave,

A better way to determine the read noise on Nikons is to use the optical black area of the raw file as Bill Claff describes.

Since you use Rawdigger, you can use it to gain access to the optical black area. In preferences check the masked pixels box and uncheck the 2x2 box. Then open the raw file and zoom to 1000% at the right edge of the image to show the masked pixels as shown here. The masked pixels are the 14 darker columns on the right.


Bill many thanks for the reference and the advice on how to use RawDigger with the masked pixels. This is all most interesting.

I didn't realise that there must be a small "bias" voltage on the raw analogue values from the sensels and that these are zeroed out for the main data. Presumably these reference black pixels are used for the purpose of this zeroing.

I had a look at the method using these reference black pixels and it gave good correlation with the SD's I calculated for the images posted above. However this method should be more accurate and it is easy to use because you can just read the SD values directly from RawDigger.

There is always something new to learn in this game, but that's what makes it fun!

Dave
Logged
Pages: 1 [2] 3   Go Up