Psal jsem si s autorem toho článku a říkal např. i toto :
FWIW, CMOS sensors don't use the same kinds of anti-blooming gates as CCD sensors do. Each pixel is independent and so you don't have the same kind of structure that leads to the need for ABG. Here are some curves on the Canons:
http://scien.stanford.edu/class/psych22 ... arity.html
Dále také napsal hezkou věc :
Have a look on page 4. The bottom row shows two CCDs that, as they are cooled, show very little change in the distribution of pixels in darks from 1m to 10m. An uncooled camera should show a decent increase in the width of the distribution (more noise, increased variance) and the distribution should shift to the right (increased intensity with increased exposure duration). There also is going to be a fixed relationship between the variance and the mean in any image (that relationship is how we can determine the system gain). Now, for the 40D, the mean is roughly constant (it drops a bit at 1600 ISO for 10m). For the 350XT, the mean goes down with increasing exposure duration. How on earth would you capture fewer photons in 10 minutes than you got in 1? Well, you didn't... Canon is shifting things.
I'm certainly not the only one to spot this and the problem then is that your flats will have one (very small) shift and your long lights will have another (much larger) shift. These shifts aren't just a simple addition / subtraction as they do seem to scale. In another article, I looked at the contrast in longer vs. shorter exposure shots and found a difference (I used the f-stop to equate the total # of photons captured). Thus, the contrast (aka amount of vignetting) will differ from this scaling. Heck, it may also differ based on the photon-transfer-curve kind of non-linearities you brought up. When I look at 10m CCD histograms, they're still showing all the data packed into a thin spike on the left of the histogram. When I look at 10m DSLR shots, they cover the whole histogram. It's quite possible that a lot more stretching is going on here in these long-exposure DSLR shots.
That makes correction of them difficult. MRAW is doing a histogram matching routine of some sort (the code is a bit tough to penetrate) to account for this. Good on 'em! While I'd not want it to do that on my cam, it seems to work wonders on the DSLRs there.
Myslím, že to mluví za vše