General debayer question

[J.D. Frey]

Ok so we have a debayer pattern like the following (I believe this is the way Red does it as well)

RGRGRGRGRGRG
GBGBGBGBGBGB
RGRGRGRGRGRG
GBGBGBGBGBGB

I'm assuming I can make a pixel with each quad

RG
GB

Can the quads "borrow" ajacent photo sites (either side or top/bottom) from each other as well to broaden color sensitivity?

RG+RG
GB+GB

yields
R GR G
G BG B

This seems like it would be a standard practice for nearest neighbors to improve color fidelity, but I guess could make the image muddy, unless you corrected for low/high luminance? Or is it a waste of processor time for little benefit?

[Anders Holck]

Debayer algorithms are much more sophisticated than simply averaging photo sites to create an output pixel: http://www.vlsi.uwindsor.ca/presenta...12-Anthony.pdf

AHD algorithm: http://www.photoactivity.com/Pagine/...03adaptive.pdf
VNG algorithm: http://scien.stanford.edu/pages/labs...ep/vargra.html

[J.D. Frey]

Awesome- thanks for the info. I didn't even know what algorithm names to start with. I've read a bunch of general stuff about bayer patterns, but have not seen any info on how the pixels are actually made. Quads seemed pretty simple, so I knew there was room in there for improvement, and I suspect the way the RED RAW/CODE wavelet compression works probably fits elegantly into a bayer pattern design.

Thanks again!