Originally posted by bdery Yes and no. What you say is true, but since for each pixels you actually have three colour channels (sometimes it's more complicated than that, with more green photosites for instance) you can remap the sensor when lowering the resolution. With a CMOS it could potentially be different since some processing can be done on the sensor. But I believe the colour channels remain separate at that level.
Is there any evidence that this "remapping" is actually happening? In other words that it is doing something special that I cannot duplicate later just by downsampling the RAW data? And how exactly would this still not be "throwing away data"? (I did not expect that to be a controversial statement since 16MP is by definition less than 24MP of data.)
Quote: It is correct to say that the sensor operates at full resolution. But the effective pixels (remapped from the bayer array) could indeed be larger.
But the actual physical pixel sites are of fixed size and with a fixed color filter (R,G, or B) and that's the entire reason noise is worse with a higher resolution sensor (that is the same size as a smaller resolution one) -- the *physical* pixels are smaller and can hold less photons. Effective pixels are just a mixture of physical pixels, right? So anything that happens after capture is software trickery. So by downsampling (essentially what this remapping would be, no?) you can have less noise, yes, but you also have less detail. You have less of everything as expected (and again, by definition). That's not the same thing as "improved noise handling".
Here's the scenario that I'm skeptical of, and I'd need proof of:
-- We take a K-3 and set it to RAW+ so it gives us the RAW file and also a JPEG-- jpegs are set for 16MP
-- We take an image (high-ISO)
-- You get the 16MP jpeg and I get the 24MP RAW file as our starting points
-- We each make the best "technical" print of that image we can -- most detail / least noise. (Still subjective, I know, but what can you do?)
-- Will your print be better than I can *possibly* make mine? (i.e. you gained something magical by using the smaller resolution jpeg file that I can't do in post-processing even with the full RAW file)
I doubt it. That's a strong claim.
But more likely is that you're not claiming something magical that can only be done at capture time, but simply that downsampling as a first step from the full resolution RAW data will give you improved noise handling. But again, even if that's true (not sure that it is given same print sizes) you're throwing away detail to do it. (Which is fine because there might be plenty to spare, but still...) I mean, that's what noise reduction software does -- smooths over noise but it also smooths over detail, and you find a balance point where the noise stops distracting you without the lack of detail annoying you too much...