John: Please put up with my obtuseness (obstinacy?) just a wee bit longer. I am well aware of the physical distinction between CCD charge transfer/summing and CMOS pixel reads. I am also in full agree with your four bullets. We're on the same page here. We may not be able to call it "resampling", but I can certainly write a software process to accomplish your bullet #3. In this case, the ADU value is four times greater, and read noise, adding in quadrature, is merely doubled. So the signal to noise ratio is doubled. That's not as good as the four-fold increase in S/N were I to use a CCD, but given that the latest CMOS sensors have far lower read noise than CCDs, at merely twice the read noise, you could still easily come out ahead.
My key goal is to shorten the total integration time required on any given object. A CCD will require four times less time on an object to achieve a given signal level if binned 2x2; imaging speed scales as inverse f-ratio squared, and as the square of the pixel size. Fat binned CCD pixels speed up imaging, although we need to be careful to avoid undersampling in the process. If we save a CMOS image at 1:1 resolution, and "bin" its pixel ADUs 2x2 per your bullet #3, the new image has one quarter the pixel count, and the ADUs of each new pixel are four times larger. So far, just like a physically binned CCD. The S/N improvement is only 2X, not 4X, but the read noise will be a lot lower with CMOS.
My point is that there is value in such software "binning" (even if we can't call it "resampling"), in that it speeds up imaging. Would you agree?
All the best,
Edited by Coconuts, 20 September 2020 - 11:39 AM.