Nikon D3 Dynamic Range test: 9 f-stops

[Guillermo Luijk]

I have done a DR subjective test on a capture using the D3. I plotted the EV zones distribution for the captured image file assuming sensor linearity. To decide whether a given f-stop is noise-free enough is up to the user and the application. I applied the criteria to be able to distinguish texture with a good degree of detail which IMO should be the best approach to define DR in a photographic context.

My final verdict is that the D3 provides at least 9 f-stops reasonably free of noise at ISO200 (lowest electronic ISO of the camera). Unluckily the 10th f-stop fell into a plain wooden door with little or no texture, so it was difficult to decide whether noise there was too high or not. IMO it was, that's why I finally admitted just 9 f-stops.

Scene:




Bright corrected for more detailed display, and plotting EV areas:




The scene extends over 11 or 12 f-stops of DR:




Finding out which ones are properly captured by the D3:






Visually the 9th f-stop (marked -8EV) still preserves wood texture, while the 10th (marked -9EV) gets already too noisy. So final usable DR is 9 f-stops which means the camera's 14-bit didn't add to much to DR (they will add quite a lot to the memory cards and hard drive manufacturing industries however).

Similar tests over my Canon 350D and Fuji S3 Pro yielded nearly 8 and 11 f-stops of DR respectively.

[Ray]

Similar tests over my Canon 350D and Fuji S3 Pro yielded nearly 8 and 11 f-stops of DR respectively.
[a href=\"index.php?act=findpost&pid=159132\"][{POST_SNAPBACK}][/a]

Guillermo.
Interesting! Have you got any results for the 5D using similar methodology?

[NikosR]

We are missing a few key things here.

First and most obvious, capture and processing method which would have a most obvious influence mainly on the noise floor but also in the highlights.

Secondly, dynamic range optimisation which would entail knowning the exposure given and its result in sensor photon-well saturation for each RGB channel.

All above can result in variations of +- 2EV in such subjective measurements of the dynamic range.

[Panopeeper]

Guillermo,

do you mind posting the raw file, so that we can take a look at it? I am interested on the level of noise in the darker areas.

[Guillermo Luijk]

Have you got any results for the 5D using similar methodology?

nope, only 350D and Fuji S3 Pro. Maybe I have some 5D RAW file in my HD but I don't remember now where it could be. I guess 5D must be about 8.5 f-stops of usable DR or so.

We are missing a few key things here.

First and most obvious, capture and processing method which would have a most obvious influence mainly on the noise floor but also in the highlights.

Secondly, dynamic range optimisation which would entail knowning the exposure given and its result in sensor photon-well saturation for each RGB channel.

All above can result in variations of +- 2EV in such subjective measurements of the dynamic range.
[{POST_SNAPBACK}][/a]

Yes you are right:

Capture method: click the shutter (I guess).
Camera: NIKON D3
ISO speed: 200
Shutter: 1/40.0 sec
Aperture: f/2.8
Focal length: 28.0 mm

Processing method: I developed the RAW file using DCRAW and camera's WB. The highlights in the upper part of the window were blown. I used mode -H 2 which guarantees not to blow any extra information that was not already blown (multipliers <=1).
The resulting histogram was linearly corrected up so the entire histogram was shifted 1/4 f-stop up to fill the empty gap on the right. This reallocated all information in the EV zones by 1/4 up providing 1/4 less DR than if we had not adjusted the histogram to the right.
I think this is the only argueable step taken and I am not 100% happy with the solution, but I think it was the most neutral decision since leaving the gap empty would mean not using the entire DR in the highlights.

This was the resulting histogram with -H 1 (pure <=1 WB multipliers, magenta highlights): no gap in the right, but usual magenta highlights:




This was the resulting histogram with -H 2 (pure <=1 WB multipliers, neutral gray highlights):



[span style=\'font-size:8pt;line-height:100%\']and after 1/4 f-stop up this is the histogram distribution finally used for the EV zones:[/span]


No noise reduction, unsharp mask, contrast or any other kind of processing applied.

Panopeeper here you are: [a href=\"http://www.guillermoluijk.com/download/nikond3_drtest.nef]nikond3_drtest.nef[/url] (24 MB)
DCRAW set black point to 0, so maybe this is in the group of cameras where black offset substraction has already been performed over the RAW data.

[bjanes]

ISO 100
[{POST_SNAPBACK}][/a]

The base ISO is 200. Was the LO 1 mode used, and how would this affect DR?

Processing method: I developed the RAW file using DCRAW and camera's WB. The highlights in the upper part of the window were blown. I used mode -H 2 which guarantees not to blow any extra information that was not already blown (multipliers <=1).
The resulting histogram was linearly corrected up so the entire histogram was shifted 1/4 f-stop up to fill the empty gap on the right. This reallocated all information in the EV zones by 1/4 up providing 1/4 less DR than if we had not adjusted the histogram to the right.
[a href=\"index.php?act=findpost&pid=159196\"][{POST_SNAPBACK}][/a]

Some readers might be disappointed with this DR result, since [a href=\"http://forums.dpreview.com/forums/read.asp?forum=1021&message=25627719]EJ Martin[/url] had reported a DR of 11.7 stops. These experiments demonstrate the importance of definitions and the experimental method when determining DR. Emil was using the engineering definition (DR = Maximum Signal/Minimum signal, usually expressed in terms of electrons: Full Well/Read Noise). Also, he was using the green channel only and performed no demosaicing.

The photographic definition uses white balanced demosaiced data, introducing more variables, and the final DR involves how much noise in the shadows is tolerable; thus, it is subjective. Imatest and the DR methodology used at DPreview are variatioins of the photographic method and involve step wedges and objective measurement of noise expressed as standard deviations, either directly or expressed in f/stops. DPreview extends the DR by using highlight recovery, but I think this is questionable. Imatest reports several DRs depending on the quality of the image.

IMHO, subjective photographic tests such as Guillermo performed are more useful to the photographer. The engineering method is more precise, but does not take into account the limitations imposed by white balance and the raw converter. If noise reduction has been applied to the image (as in many in camera JPEGs), another variable has been added.

DCRAW set black point to 0, so maybe this is in the group of cameras where black offset substraction has already been performed over the RAW data.
[a href=\"index.php?act=findpost&pid=159196\"][{POST_SNAPBACK}][/a]

From what I have read, the Nikon applies no offset and merely clips the black point. Many regard this behavior as undesirable.

[Panopeeper]

Guillermo,

I looked at the raw image; I (subjectively) agree with your (subjective) conclusion, that the 9th stop is still usable. The ninth stop is the range of pixel values between 30 and 60. Under that it is very noisy.

Btw, it's interesting, that the clipping level for red is 15740, for blue 15760, but for half of the greens it is 16260 and for the other half 16383. I saw raw files from several cameras exhibiting such behaviour; I wonder, what the reason is for that. Perhaps the raw image we get to see is not raw enough any more.

[Guillermo Luijk]

The base ISO is 200. Was the LO 1 mode used, and how would this affect DR?

Sorry, shot was taken at ISO200. I am used to my 350D's ISO100.

I (subjectively) agree with your (subjective) conclusion, that the 9th stop is still usable. The ninth stop is the range of pixel values between 30 and 60. Under that it is very noisy.

I'm glad to hear that. BTW, new test, now for the Sony A700. It beats the Nikon!:
Sony A700 Dynamic Range Test: 9.5 f-stops

[douglasf13]

Sorry, shot was taken at ISO200. I am used to my 350D's ISO100.
I'm glad to hear that. BTW, new test, now for the Sony A700. It beats the Nikon!:
Sony A700 Dynamic Range Test: 9.5 f-stops
[{POST_SNAPBACK}][/a]

You guys can check this out.  DIWA tests of both cameras.

[a href=\"http://www.diwa-labs.com/wip4/test_result.epl]http://www.diwa-labs.com/wip4/test_result.epl[/url]