Pixel size does not impact DR?

[dreed]

As per this graph, the Nikon D600 and D800 have approximately the same DR:
http://home.comcast.net/~NikonD70/Charts/PDR.htm#EOS%205D%20Mark%20III,D600,D800

This seems to go against the line of thinking that associates bigger pixels with more DR.

[jeremypayne]

"D600 data is preliminary and incomplete."

Wondering what this means exactly ...

[bjanes]

As per this graph, the Nikon D600 and D800 have approximately the same DR:
http://home.comcast.net/~NikonD70/Charts/PDR.htm#EOS%205D%20Mark%20III,D600,D800

This seems to go against the line of thinking that associates bigger pixels with more DR.

The total sensor area is the most important factor, since it determines the total number of photons collected. If tghe sensor size is held constant (both of these cameras are full frame), the larger pixel camera will have better per pixel SNR and DR. However, if one prints at the same size, the smaller pixel camera will gain from pixel averaging as the image is downsampled to the print resolution. This is the basis for DXO's screen and print DR values.


See this DXO whitepaper for an analysis.

Regards,

Bill

[mac_paolo]

This seems to go against the line of thinking that associates bigger pixels with more DR.

Marginally. The most important factor is the sensor technology. For the state of the art two different pixel pitch sensor have different maximum achievable detail, but DR and color sensitivity should be close if not almost equal.
It wouldn't surprise me that much if D800 and D600 share some of their characteristic. It would be more expensive for Nikon (or any camera manufacturer) to develop a radically different sensor for """similar""" (they are not night and day at all) camera bodies.

[Graystar]

This seems to go against the line of thinking that associates bigger pixels with more DR.

Well, such thinking was always wrong to begin with, but anyways...

Forget about DR.  All cameras today are going to have nearly 12 stops of DR at the very least, so it's pretty meaningless now.  In today's cameras, DR is a result of processing decisions more than anything else.  The reason this is so is that DR is limited by the bit resolution of the camera's ADC.  If you could measure the analog DR of the sensor, it would likely be larger than what DxOMark is reporting.  This is why a Canon Powershot S100 can have nearly as much DR as a Canon 5DMIII.

The measure you really want is SNR.  Look at the SNR and there's a huge difference between the two cameras.  The SNR tells the real story (which is, in this example, that the 5DMIII has about 3.5 stops less noise than the S100.)

[deejjjaaaa]

"D600 data is preliminary and incomplete."

Wondering what this means exactly ...

as far as I understand people supply William Claff with raw files shot per his instructions so that he can do a more precise measurements... instructions and software can be downloaded from his website... if you have a camera not listed on his website you can supply him with such data and contribute to the camera's database (not only Nikon and Canon), otherwise he uses some publicly available raws (imaging resources like) and hence he probably notes that his calculations are preliminary, etc.

[bjanes]

Well, such thinking was always wrong to begin with, but anyways...

Forget about DR.  All cameras today are going to have nearly 12 stops of DR at the very least, so it's pretty meaningless now.  In today's cameras, DR is a result of processing decisions more than anything else.  The reason this is so is that DR is limited by the bit resolution of the camera's ADC.  If you could measure the analog DR of the sensor, it would likely be larger than what DxOMark is reporting.  This is why a Canon Powershot S100 can have nearly as much DR as a Canon 5DMIII.

The measure you really want is SNR.  Look at the SNR and there's a huge difference between the two cameras.  The SNR tells the real story (which is, in this example, that the 5DMIII has about 3.5 stops less noise than the S100.)

Graystar is entirely correct in so far as he goes, but there is more to the story. With linear capture the engineering DR (full well capacity/read noise) is limited by the bit depth of the ADC. A bit depth of 12 imposes a DR limit of 12 stops. However, the DR of small pixel cameras falls off dramatically at higher ISOs as shown in Figure 7 of Roger Clark's post. The denominator in the engineering DR equation is the read noise, which is the noise with the lens cap on the camera. This level of illumination is not useful for photographic purposes. Bill Claf reports photographic DR (according to his criteria), which is more useful.

Shot noise (photon counting noise) is predominant over most of the useful photographic DR and larger pixels will collect more photons and have a better SNR. A full SNR plot as given by DXO allows calculation of the DR for any noise floor (see Emil Martinec).

Regards,

Bill