Ad
Ad
Ad
Pages: « 1 [2] 3 »   Bottom of Page
Print
Author Topic: Dynamic range and DxO  (Read 23191 times)
bjanes
Sr. Member
****
Offline Offline

Posts: 2866



« Reply #20 on: March 10, 2010, 08:02:41 AM »
ReplyReply

Quote from: thierrylegros396
Sound just like in the audio world with a lot of subjectivism, but also real unexplainable differences      

Have a Nice Day.

Thierry
Since you bring up the analogy with audio, you might be interested in The great harmonic distortion scam by Norman Koren. Total Harmonic Distortion at rated power was used to judge amplifiers and some solid state systems did quite well with this measurement and had similar or better specs to tube amplifiers. However, transients which are usually present in music led to clipping and massive distortion and harsh appearing sound.

Norman concludes, "That brings us to the ultimate audio paradox: In a well-designed high fidelity power amplifier, the lower the total harmonic distortion at rated power, the harder the clipping; hence the worse the sound.".

Just as in audio, in digital we have to be sure that what we are measuring correlates with perceived quality of reproduction.

Bill
« Last Edit: March 10, 2010, 08:03:35 AM by bjanes » Logged
eleanorbrown
Sr. Member
****
Offline Offline

Posts: 632


WWW
« Reply #21 on: March 10, 2010, 08:38:19 AM »
ReplyReply

Michael, as an aside here, I would be interested to hear your take where the Leica M9 at base iso (ie: 160), fits into the dynamic range between the 65+ and say the DSLR's at their base iso.?  Closer to the digital back or the DSLR's.  Eleanor


Quote from: michael
In case anyone hasn't seen it, the following was added to the review yesterday...




The issue of the differences between medium format and 35mm is a fascinating one though, and Mark and I intend on pursuing it in greater depth in the days ahead, and with more rigour than with a throw-away line or two.[/i]
Logged

ejmartin
Sr. Member
****
Offline Offline

Posts: 575


« Reply #22 on: March 10, 2010, 11:08:48 AM »
ReplyReply

Quote from: ErikKaffehr
I have another question, why are MFDBs relatively poor performers at high ISO? Whatever advantage they have they also be able to keep it at higher ISO.

The factors I see are:

Most MFDBs are said not to have variable pre-amps, so the ISO settings above "base ISO" are just "fake ISO". This seems not be the case with then new P65 plus, however.

MFDBs normally don't have microlenses. Microlenses don't play well with lens shifts to my understanding.

The CGA may use narrower filters. This could result in more saturated colors, but not necessarily better colors.

Best regards
Erik

OK, another little tutorial on how to read DxO data:

DR is limited by the weakest link in the signal processing chain (much like MTF is limited by the poorest performing component in the optical path).  The sensor has a DR, the ISO amplifier has a DR, and the A/D converter (ADC) has a DR.  Ideally, the electronic components downstream of the photosites in the signal processing chain would have a DR as large or larger than the DR of the photosite array, otherwise there is information captured by the photosites that is lost along the way by the noise contributed by these other components.  When the ADC's DR is smaller than the photosite DR, one has to choose what part of the photosite's range is to be delivered to the raw data.  One achieves this by varying the ISO gain -- larger gain amplifies the lower end of the photosite DR to fit it into the DR window of the ADC, less amplifiation fits the upper end in.

Thus if we look at the DxO (engineering) DR plot for say the Nikon D3, D3s, and P65+ ("screen" tab, so pixel DR)

we see that the the D3s has a very long flat part at low ISO.  This is occurring because the amp/ADC DR is about 11.6 stops, and is the limiting factor in the low ISO DR.  Increasing the ISO slides that DR towards the lower end of the photosite array's DR, and one hits that lower end about ISO 3200 or so, at which point the DR drops by about a stop for every increase in ISO (because increasing the ISO by one stop pushes one more stop of highlights past the saturation point of the ADC's range).  In fact we can deduce from the fact that the D3s DR at ISO 12800 is a bit over 8.2 stops (according to DxO) that the photosite array must have a DR of over 14 stops (since by ISO 12800 one has thrown away the top 6 stops of range of ISO 200 in terms of absolute light intensity, and 8.2+6=14.2); the low ISO engineering DR of the D3s is severely limited by the amp/ADC DR.  (BTW, the kink in the P65+ plot is due to its use of pixel-binning starting at ISO 800.)

On the other hand, the D3x and P65+ have a DR that starts dropping by one stop per ISO almost immediately.  This indicates that the DR of the amp/ADC is about equal to the DR of the photosite array, so the components are well matched (for the D3x, this is likely related to the massively parallel on-chip ADC of the sensor architecture).  This is what we would want in an ideal world -- none of the sensor's range is being compromised by downstream electronics.  IF the downstream electronics has a DR that exceeds the sensor sufficiently, then there is no reason to offer variable ISO gain in the camera -- you've already captured the entire sensor range at base ISO, and can achieve higher ISO by simply telling the raw converter to push process according to metering.  In a well-designed camera, ISO should be metadata -- and it is on many MFDB's.

A fairer comparison for ISO performance normalizes for the number of pixels in the frame; if we do that (DxO's "print" tab), we find

and the P65+ has with their normalization about 13 stops engineering DR in a fixed size print.  The D3s/D3x are in that ballpark at their base ISO (the D3s a bit less due to the poor amp/ADC DR, the D3x a bit more because more of its photosite DR is realized), but their base ISO is higher than the P65+, which means their sensors are more efficient at converting incident light to sensor signal (the so-called quantum efficiency).  That means they carry more of that range into high ISO's and achieve better low-light performance.

EDIT: The DSLR advantage at high ISO is rather mostly due to better read noise behavior, not just QE (the lower QE of the MFDB is somewhat offset by the larger sensor area).  Perhaps it is better to say that the DR of the photosite array is larger, and by the time we reach high ISO the amp/ADC DR is no longer a limiting factor.  The DSLR's simply have better photosite read noise -- about 4.5 electrons for the D3x, about 2.5 electrons for the D3s, compared to about 16 electrons for the P65+.  Looking at the graphs again, it is this read noise difference that accounts for the bulk of the over two stop difference in engineering DR between the Phase back and the D3x, and the over three stop difference with the D3s, at high ISO.  Again, though, all of this comes at the shadow end; all these cameras collect about the same number of photons over the frame, so the difference is not as great when we impose stricter minimum S/N criteria.
« Last Edit: March 11, 2010, 09:36:27 AM by ejmartin » Logged

emil
ErikKaffehr
Sr. Member
****
Online Online

Posts: 7915


WWW
« Reply #23 on: March 10, 2010, 01:32:18 PM »
ReplyReply

Emil,

Thank's a lot. This answers some of the questions I had.

I'm really curious about what Mark and Michael will find out comparing MFDBs and DSLRs. For me it's not about finding the winner but understanding the differences.

Best regards
Erik



Quote from: ejmartin
OK, another little tutorial on how to read DxO data:

DR is limited by the weakest link in the signal processing chain (much like MTF is limited by the poorest performing component in the optical path).  The sensor has a DR, the ISO amplifier has a DR, and the A/D converter (ADC) has a DR.  Ideally, the electronic components downstream of the photosites in the signal processing chain would have a DR as large or larger than the DR of the photosite array, otherwise there is information captured by the photosites that is lost along the way by the noise contributed by these other components.  When the ADC's DR is smaller than the photosite DR, one has to choose what part of the photosite's range is to be delivered to the raw data.  One achieves this by varying the ISO gain -- larger gain amplifies the lower end of the photosite DR to fit it into the DR window of the ADC, less amplifiation fits the upper end in.

Thus if we look at the DxO (engineering) DR plot for say the Nikon D3, D3s, and P65+ ("screen" tab, so pixel DR)

we see that the the D3s has a very long flat part at low ISO.  This is occurring because the amp/ADC DR is about 11.6 stops, and is the limiting factor in the low ISO DR.  Increasing the ISO slides that DR towards the lower end of the photosite array's DR, and one hits that lower end about ISO 3200 or so, at which point the DR drops by about a stop for every increase in ISO (because increasing the ISO by one stop pushes one more stop of highlights past the saturation point of the ADC's range).  In fact we can deduce from the fact that the D3s DR at ISO 12800 is a bit over 8.2 stops (according to DxO) that the photosite array must have a DR of over 14 stops (since by ISO 12800 one has thrown away the top 6 stops of range of ISO 200 in terms of absolute light intensity, and 8.2+6=14.2); the low ISO engineering DR of the D3s is severely limited by the amp/ADC DR.  (BTW, the kink in the P65+ plot is due to its use of pixel-binning starting at ISO 800.)

On the other hand, the D3x and P65+ have a DR that starts dropping by one stop per ISO almost immediately.  This indicates that the DR of the amp/ADC is about equal to the DR of the photosite array, so the components are well matched (for the D3x, this is likely related to the massively parallel on-chip ADC of the sensor architecture).  This is what we would want in an ideal world -- none of the sensor's range is being compromised by downstream electronics.  IF the downstream electronics has a DR that exceeds the sensor sufficiently, then there is no reason to offer variable ISO gain in the camera -- you've already captured the entire sensor range at base ISO, and can achieve higher ISO by simply telling the raw converter to push process according to metering.  In a well-designed camera, ISO should be metadata -- and it is on many MFDB's.

A fairer comparison for ISO performance normalizes for the number of pixels in the frame; if we do that (DxO's "print" tab), we find

and the P65+ has with their normalization about 13 stops engineering DR in a fixed size print.  The D3s/D3x are in that ballpark at their base ISO (the D3s a bit less due to the poor amp/ADC DR, the D3x a bit more because more of its photosite DR is realized), but their base ISO is higher than the P65+, which means their sensors are more efficient at converting incident light to sensor signal (the so-called quantum efficiency).  That means they carry more of that range into high ISO's and achieve better low-light performance.
Logged

ErikKaffehr
Sr. Member
****
Online Online

Posts: 7915


WWW
« Reply #24 on: March 10, 2010, 02:18:22 PM »
ReplyReply

-- Deleted --
I posted some information I should not have posted. Sorry!
« Last Edit: March 10, 2010, 02:31:13 PM by ErikKaffehr » Logged

deejjjaaaa
Sr. Member
****
Offline Offline

Posts: 1082



« Reply #25 on: March 10, 2010, 03:33:15 PM »
ReplyReply

Quote from: ErikKaffehr
Emil,

Thank's a lot. This answers some of the questions I had.

I'm really curious about what Mark and Michael will find out comparing MFDBs and DSLRs.

whatever they will find out, I bet they will not post any raw files, so that anybody can get CaptureOne (non owners can download a trial) and compare themselves.
Logged
ErikKaffehr
Sr. Member
****
Online Online

Posts: 7915


WWW
« Reply #26 on: March 10, 2010, 03:49:20 PM »
ReplyReply

Hi,

Michael Reichmann made a test together with Bill Atkinson and Charlie Cramer a couple of years ago and offered the raw images on a DVD. Try to be a bit positive!

Best regards
Erik

Quote from: deja
whatever they will find out, I bet they will not post any raw files, so that anybody can get CaptureOne (non owners can download a trial) and compare themselves.
« Last Edit: March 10, 2010, 03:52:51 PM by ErikKaffehr » Logged

BartvanderWolf
Sr. Member
****
Offline Offline

Posts: 3876


« Reply #27 on: March 10, 2010, 04:21:35 PM »
ReplyReply

Quote from: ejmartin
OK, another little tutorial on how to read DxO data:[...]

Hi Emil,

Just wanted to say, excellent explanation, again, thanks.

Cheers,
Bart
Logged
bradleygibson
Sr. Member
****
Offline Offline

Posts: 829


WWW
« Reply #28 on: March 10, 2010, 08:56:52 PM »
ReplyReply

Thank you, Emil.  Great stuff.
Logged

douglasf13
Sr. Member
****
Offline Offline

Posts: 547


« Reply #29 on: March 10, 2010, 11:17:08 PM »
ReplyReply

Quote from: ErikKaffehr
Hi,

What about lenses, flare level and so on? DSLR lenses used to be complex often zooms. A simpler lens with fewer air/glass surfaces would have less flare. Lenses with moving internal lens groups probably cannot be as efficient in repressing internal reflections as lenses with all lens groups fixed.

Best regards
Erik

 This is what I want to here more about.  Iliah Borg claims that his D3x's DR is limited not by the sensor, but, rather by lenses and the flare/internal reflections associated with them.  I believe he states that even the simplest prime with few elements only allows 11 stops or so.
Logged
DarkPenguin
Guest
« Reply #30 on: March 10, 2010, 11:54:02 PM »
ReplyReply

Quote from: douglasf13
This is what I want to here more about.  Iliah Borg claims that his D3x's DR is limited not by the sensor, but, rather by lenses and the flare/internal reflections associated with them.  I believe he states that even the simplest prime with few elements only allows 11 stops or so.

http://forums.dpreview.com/forums/read.asp...essage=34759823
Logged
deejjjaaaa
Sr. Member
****
Offline Offline

Posts: 1082



« Reply #31 on: March 11, 2010, 10:09:40 AM »
ReplyReply

Quote from: ErikKaffehr
Hi,

Michael Reichmann made a test together with Bill Atkinson and Charlie Cramer a couple of years ago and offered the raw images on a DVD. Try to be a bit positive!

Best regards
Erik

so what is his problem to post couple of raw images online if he already has them and they prove "6 stops" advantage   ... may be they don't really, huh ?
Logged
bjanes
Sr. Member
****
Offline Offline

Posts: 2866



« Reply #32 on: March 11, 2010, 10:42:20 AM »
ReplyReply

Quote from: douglasf13
This is what I want to here more about.  Iliah Borg claims that his D3x's DR is limited not by the sensor, but, rather by lenses and the flare/internal reflections associated with them.  I believe he states that even the simplest prime with few elements only allows 11 stops or so.
Iliah is a smart guy and a careful worker. Norman Koren discusses veiling glare (flare) and it can be measured with Imatest. One of the best lenses he tested had a veiling glare of 0.3%. For a linear file, veiling glare = V = L(black object) / L(white surface), where L is the illuminance at the sensor. If you define DR as the ratio of the lightest to the darkest tone, then an image taken with a lens having veiling glare of 0.3% would have a DR of 100/0.3 = 8.4 stops.

Users of Imatest to determine the DR of cameras with a Stouffer wedge know that it is important to mask off the target from the background of the lightbox so as to avoid flare as much as possible. Here is such a target photographed with and without masking with the Nikon D3 and the 60 mm f/2.8 AFS which has the Nano multicoating and relatively low flare. In the unmasked shot, the shadow values are filled in by flare light.

[attachment=20823:Masking.png]

Using the formula for veiling glare, one can calculate the luminances of the Stouffer wedge with and without flare light as shown in this Excel spread sheet.

[attachment=20824:ExcelGlare.png]

Without any flare light, the ratios between adjacent steps are constant, but with flare light the steps are still present but less differentiated. Imatest determines DR according to the number of steps detected. Even with flare light, the target would still have 40 steps (13 stops) but their detection would cease when the noise of the camera overwhelms the differences in the progressively smaller steps.
« Last Edit: March 11, 2010, 10:45:03 AM by bjanes » Logged
ErikKaffehr
Sr. Member
****
Online Online

Posts: 7915


WWW
« Reply #33 on: March 11, 2010, 12:02:53 PM »
ReplyReply

Hi,

This was a couple of years ago and was about resolution and not DR. I have a DVD with a lot of raw images from all cameras.

Regarding the new "test" they plan we have to wait and see.

Best regards
Erik


Quote from: deja
so what is his problem to post couple of raw images online if he already has them and they prove "6 stops" advantage   ... may be they don't really, huh ?
Logged

deejjjaaaa
Sr. Member
****
Offline Offline

Posts: 1082



« Reply #34 on: March 11, 2010, 03:41:09 PM »
ReplyReply

Quote from: ErikKaffehr
Hi,

This was a couple of years ago and was about resolution and not DR.

but this is about DR... did anybody dispute resolution here ?
Logged
ErikKaffehr
Sr. Member
****
Online Online

Posts: 7915


WWW
« Reply #35 on: March 11, 2010, 10:30:20 PM »
ReplyReply

What I have written that LuLa published raw files from a similar test in the past. If they do it now we don't know, let's hope for it. In my view they should charge a reasonable fee for it, because it takes a lot of bandwith.

To evaluate DR you need special arrangements. Photographing a "Stuffer Edge" is OK, but I guess that Mark and Michael want to make a more subject oriented demonstration.

Best regards
Erik


Quote from: deja
but this is about DR... did anybody dispute resolution here ?
« Last Edit: March 11, 2010, 10:33:37 PM by ErikKaffehr » Logged

Jack Flesher
Sr. Member
****
Offline Offline

Posts: 2595



WWW
« Reply #36 on: March 12, 2010, 09:47:18 AM »
ReplyReply

Perhaps the more important factor for photographic artists when discussing any given camera's DR, is not so much how much total DR is there to begin with, but what you can do with it before you lay it down in an image...

Cheers,
Logged

deejjjaaaa
Sr. Member
****
Offline Offline

Posts: 1082



« Reply #37 on: March 12, 2010, 06:57:58 PM »
ReplyReply

Quote from: ErikKaffehr
What I have written that LuLa published raw files from a similar test in the past. If they do it now we don't know, let's hope for it. In my view they should charge a reasonable fee for it, because it takes a lot of bandwith.

imaging resources host hunderds of raw files w/o issues and we just need 2 files...  and to assume that professional photog like the site owner somehow does not have the raw files from the important test in the past while he has means to maintain his vast library of pictures is kind of unbelievable, I am sorry.


Logged
Ray
Sr. Member
****
Offline Offline

Posts: 8939


« Reply #38 on: March 12, 2010, 07:25:52 PM »
ReplyReply

Quote from: Jack Flesher
Perhaps the more important factor for photographic artists when discussing any given camera's DR, is not so much how much total DR is there to begin with, but what you can do with it before you lay it down in an image...

Cheers,

Jack,
I presume you are referring to practices like applying a neutral density gel to windows to bring the SBR of the view outside the window closer to the SBR of the scene inside the room, and of course, using fill flash to illuminate shadows when distances are not too great, or strategically placing artificial lighting where practicable. Is that right?
Logged
feppe
Sr. Member
****
Offline Offline

Posts: 2909

Oh this shows up in here!


WWW
« Reply #39 on: March 12, 2010, 07:37:09 PM »
ReplyReply

Quote from: deja
imaging resources host hunderds of raw files w/o issues and we just need 2 files...  and to assume that professional photog like the site owner somehow does not have the raw files from the important test in the past while he has means to maintain his vast library of pictures is kind of unbelievable, I am sorry.

If you're referring to free image hosts, I'm not aware of any which host RAWs weighing tens of megs. There are several free file hosts, but those have pretty onerous upload and download limitations, and have quite unfriendly UIs.

If necessary, I can help since my web server has 300 gigs of bandwidth per month (5,000 RAW files assuming 60MB each) without me having to add more.
Logged

Pages: « 1 [2] 3 »   Top of Page
Print
Jump to:  

Ad
Ad
Ad