Down sampling does reduce noise... it's not a theory. It is used in many fields including still and motion picture. Many features are shot at a higher res than distribution, but for
celluloid and digital distribution.
I think you misattributed the word "theoretical" in that sentence. I was perhaps not clear enough.
Down sampling does reduce apparent noise. In theory this reduction is around one stop.
DXO graphs are a valid source. Even Phase One refers customers to DXO. I remember getting emails when the IQ180 report was published and Phase One refers to
DXO on it's website:
http://www.phaseone.com/en/Camera-Systems/Camera-Bodies/Reviews.aspx
Anyway here are some images of sensor plus from a P65+
ISO 800
ISO3200
800 ISO scaled over the 3200 ISO
While the noise appears similar at the same magnification it is far better when the ISO 800 is scaled and it looks better than
one stop to me.
We all know that noise increases quite dramatically with MF CCD sensors and not in a linear manor once you go over 400/800.
Most likely the noise increase would have been worse even with down sampling with an IQ180 80mp ISO 1600 capture.
So the choice of using binning is a good move or else they would not have done it. But it's worth keeping in mind that it
is to address the limitations of CCD sensors when it comes to high ISO.
The advantage though of having sensor plus is worth noting.
Slightly faster frame rate 0.7 fps (0.9 fps in Sensor+ ) for the IQ180. That may not seem like a big difference, but for such a slow camera it is
a 28% speed increase. There are also work flow advantages. If you are going to down sample it can be an advantage to have the camera do it at the get go.
However it's worth considering if it's worth the extra cost as sensor plus is far from free. The Leaf backs don't offer sensor plus,
but shoot at full MP count at ISO 800 and can be scaled down. The difference in price is quite significant especially if
you keep in mind that you can get the Leaf form B and H as well as Adorama.
I think this reinforces the argument for CMOS for MF.
Lol, yes fred, Phase One's marketing department, like every other marketing department will reference and link to anyone who says nice things about them. If President Obama made an offhand remark that he thought Phase One backs produced great colors I'm sure Phase One's marketing would send out an email with that quote - that wouldn't make Obama any more authoritative on color rendering than joe schmoe.
Citing P1 Marketing to show that dXo is a point of reference with broadly applicable meaning is silly. dXo have a very specific set of tests that tell you a very specific thing. They provide no useful guidance on very important nuances like what KIND of noise, how pretty the grain of that noise is, what the impact on the noise will be of using the native vs. generic raw processor, the impact of ambient temperature, length of exposure etc. If we printed 1's and 0's then they'd be everything you need to know about an image. But we don't. We print colors, and textures, and details, and tonality, and transitions and a dozen other aesthetic attributes which can only be loosely (and at great effort) be described or analyzed mathematically. The only way to really see what kind of picture a camera will take is to take a picture with that camera. Everything else is armchair analytics.
As I've said a dozen times, the KIND of noise matters even more in most cases than the amount of noise. Banding, color crossover, blotchiness, low frequency color wonking, and erratic non-gaussianly distributed noise will ruin the aesthetics of an image far before a finely distributed film-like evenly structured homogeneously colored grain will, regardless of what their resulting respective SNRs (signal to noise ratios) are.
We understood this in the days of film. No one ranted about scientific charts of how grainy this emulsion was or that emulsion was (or maybe they did, I was too young to have noticed); they shared and looked at photos with different emulsions to see what kind/amount/look of grain it produced. They looked at pictures.
As a case in point: those images you posted were clearly processed in Capture One v6 or LR/ACR (or had awful settings applied in v7). The quality of higher ISO processing for Phase/Leaf files is remarkably better in version 7. The color bleed you see, the lack of color subtlety, and the poorly shaped grain structure are simply in your examples are absent in v7. Check out an
example on our facebook page.
I'm outta here for a while. This is tiring. I'll see you all in the new year.