Is it in any way possible to bin pixels when using Bayer filter?
I think there where some discussion around that for some mobile phone sensor a while ago.
There's no nice way to do this for color (you're asking a resampling purist). Try it yourself. Take the pattern:
RGRGRGRG...
GBGBGBGB...
RGRGRGRG...
GBGBGBGB...
RGRGRGRG...
GBGBGBGB...
....
... then, draw squares surrounding 2x2 tiles of blue and also for red, and then on 2x2 diamonds for the greens. Now matter how you decide where to make the groups, you get a lopsided, asymmetrical mess, and there is no way to demosaic it without damaging it further, unless you demosaic back to the original full resolution, because the weighted centers of these tiles fall in a grid at the original resolution. Even before you draw groups in the green channel, it is obvious that the red and blue channels are going to have weighted centers one diagonal original pixel apart.
This wouldn't matter much if the pixel density was high enough to oversample the optics, to begin with, but it's not, and people like their acuity, and you're going to lose it like that. You have to blur either the red or blue channel significantly to get the image to an output pixel count 1/4 the original. One would hope that binning 4 pixels into one and quartering the MPs would result in an output that was more like a Foveon with three colors at each pixel, but this would be nothing like that, really, as colors would be spread by the devil that is resampling.
For monochrome, you could get away with ignoring the CFA and pretend that the sensor was monochrome, and bin 2x2 tiles of pixels, but the results will have the underlying 3 color bands shifted on a diagonal with green in the center. You won't see this as color fringing, but it can express as a diagonal directional blur.
