D3 sensor efficiency

Started Nov 18, 2007 | Discussions thread
OP ejmartin Veteran Member • Posts: 6,274
Re: Maximum black with reflective target

Bill Janes wrote:


Thanks for the explanation, which I find very interesting. The
illumination in the post you quoted was about 60 in 14 bit ADUs, in
line with what I thought it might be. Since Nikon clips the blacks,
is this background illumination akin to the bias that Canon adds to
prevent this problem?

No, it is different. For Canon, black, meaning zero illumination of the sensor, is shifted to a nonzero raw level. For Nikon, zero illumination of the sensor is zero raw level. Modulo the spread around those values due to noise, of course.

I see that you did obtain blackframes, and I presume they were
similar to what you obtained by extrapolation. Which method is
preferable? Certainly one does not need a regression such as Emil
Martinec performed.

Um, he is me. We did not obtain blackframes, we found a raw file that had uniform OOF patches of very low illumination. These will have a very slow drift upward away from zero due to the addition of photon noise; by extrapolating back to zero you recover the read noise at zero illumination. If you measure noise above black it will not be pure read noise but will have a slight contamination of photon noise and you will overestimate it. If you measure read noise at black you will underestimate it because of the clipping to a half-gaussian, which has a smaller std dev than the full gaussian by a factor 1.66. By extrapolating as I (he?) did you avoid the clipping and remove any contribution of photon noise.

BTW, I see that he had some trouble with Iris reading the TIFFs from
DCRaw. ImageJ, the freeware program from the National Institutes of
Health can read and analyze 16 bit TIFFs (but not the compressed ones
output by IRIS) and also FITs that Iris outputs. Roger Clark likes
ImagesPlus. Which do you prefer for this purpose?

I'm a Mac guy and ImagesPlus is out (barring the use of Parallels or VMWare; I tried it under crossover and it was excruciatingly slow). ImageJ is nice for some purposes, but I prefer IRIS for the analysis of raw files. In particular ImageJ can't extract color channels G1,B,R,G2 from a CFA pattern AFAIK.

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