Originally posted by Rondec Pixel shift does reduce noise -- quite a bit and motion correction algorithms take care of moving objects in the screen, but it is still somewhat limited in its usefulness from that standpoint.
Pixel shift is great, when it can be applied. It reduces noise in two ways I can discern: by taking 4 pictures, it does some amount of averaging on non-reproducible types of noise, and, more subtly, by doing it for the different color pixels it could to some extent reduce bad colour reconstruction due to the different sensitivities and nonlinearities of the different color pixels, although I think this second effect is probably already very carefully taken care of if normal images.
Originally posted by Rondec As far as your comments about noise reduction. The question to me isn't whether the accelerator chip effects the overall signal. The question is if the Pentax solution (a) does a better job at noise reduction than 95 percent of us could do with software available to us (in my case it does) and (b) whether the signal reduction is noticeable? This second question is not asking if signal is effected, but whether you see blotchiness, over smoothing of images or any of the tell tale signs that noise reduction is in play. Personally, I don't see it and the results are excellent, but I can understand if just the idea of any noise reduction occurring in the background upsets you.
I was not particularly talking about the AU actually, but responding to the post I quoted that seemed to indicate that we can simply just have a total signal consisting of total = signal + noise, and then get the 'pure' signal by doing signal = total - noise. Which would be nice, but unfortunately life is not that easy.
Like most people here, I am very interested to find out what the AU does, and what type of algorithm it uses. It might be very difficult to find that out. I am also very interested in your (a) and (b) questions. As it is, I do not (yet) have any of the newer cameras, but one of the reasons I am stalking the forums is to see what people's experience is with the new accelerator unit, and if they perceive any loss of detail. In principle, an algorithm can be devised that recognizes noise as well as the human eye does, but such an algorithm would be much more complicated and computationally expensive than the very simple approaches I pointed out before. As far as I know, no existing algorithm so far gets close. It is very likely that the AU does do something non-trivial if it reduces noise without an easily observable loss of detail. I guess RAW comparisons of a K-3ii and a KP might be interesting to pixel-peep at.