D3 sensor efficiency

Started Nov 18, 2007 | Discussions thread
DuncanM1 Contributing Member • Posts: 817
Re: Some thoughts on this

Yes, I was wondering what advantage Nikon/Pentax/Sony gain from using a 0 offset and removing negative values? If Canon is adding the bias and not removing it, then it would narrow the range of ADU values and increase quantization errors.

However, when you think about it, why bother including pixel values that you know contain nothing but noise? If this does cause a loss of DR, then Nikon et all, are already being "penalized" when, for example, Imatest measures the DR of the camera.

In any event, I measured a RMS noise of 14e for the D40X by subtracting bias frame pairs, so if we increase it by 1.66 we would get 23.5e.

I measured about 8e on my D50, using the above method, so that brings it to 13.5e which is in complete agreement with Clark's data, if we correct by 1.66.

On a K10D, I measured RMS noise of .9 ADU, so assuming a similar value for a D200, we would get 7e so a final value of 10e, as measured by Clark seems plausible, and a range of 10 to 14e seems likely to cover the range of possible values.

I do know that Canon clips their lower ADUs during long exposures:

350XT, 30min exposure, ISO200:

Red layer
Mean: 242.7 Median: 241
Sigma: 19.8
Maxi.: 4095.0 Mini.: 0.0
Green layer
Mean: 244.6 Median: 243
Sigma: 21.7
Maxi.: 4095.0 Mini.: 0.0
Blue layer
Mean: 246.5 Median: 244
Sigma: 20.9
Maxi.: 4095.0 Mini.: 0.0

Bias, same camera, and ISO:

Red layer
Mean: 255.4 Median: 256
Sigma: 12.4
Maxi.: 271.0 Mini.: 0.0
Green layer
Mean: 255.7 Median: 256
Sigma: 12.4
Maxi.: 295.0 Mini.: 0.0
Blue layer
Mean: 256.2 Median: 257
Sigma: 12.4
Maxi.: 273.0 Mini.: 0.0

the min values of 0 are probably because IRIS is reading a masked portion of the frame, and including those values in the summary.

My D50 has an Darkframe has an absolutely linear increase in ADU values up to the max of 30minutes, (measured in Mode3).


Hello Emil,

Once again, great and interesting work. If you like, a more
descriptive term (than "gain") for the e- DN that you measure is the
"camera gain constant" (from the work of J. Janesick).

Also, I'm wondering why Nikon would do this clipping. As a
scientist, I would expect them to do the logical thing and add the
bias, etc... (the astrophotographers expecially appreciate this).
But for general photography, maybe they think that it's ok to have
zero bias and this is "good enough". I don't know, how do those dark
areas "look"? And, if objectionable by enough people, could a change
in bias levels be done in a firmware update?

And also, when Canon does add this bias, is there some proof
somewhere (maybe already done by you or others, and I just haven't
seen it) that they subtract it off again before doing the A/D?

Take care,


Post (hide subjects) Posted by
Keyboard shortcuts:
FForum PPrevious NNext WNext unread UUpvote SSubscribe RReply QQuote BBookmark MMy threads
Color scheme? Blue / Yellow