One of the green channels is more sensitive than the other (it has a higher spectral transmissivity - it lets more wavelengths through). It\'s also expanded towards the red channel by quite a lot.
So to get maximum colour accuracy, you need to demosaic the 7D as \"four different colours\", it\'s not R+G+G+B.... It\'s more like Red+Yellowgreen+Green+Blue.
What is the evidence for this (rather than, say, mismatched column gains on particular copies)? Do you for example see a difference between the spectral response of the two green channels in the IR test files (eg the one that includes a colorchecker chart)?
I\'ve added an example of how this effects colours that are somewhere between yellowish green and warm red. This is the EXACT same raw, with EXACTLY the same settings of sharpening, converted in EXACTLY the same raw engine (my own MatLab development). The only thing that differs is that one is de-mosaiced with an algorithm that supposes that all four colours are totally different from eachother, normal demosaicing usually supposes that the two greens in each RGGB-group are the same colour.
Such a demosaic scheme gives up a huge amount in resolution -- if all four subarrays are assumed independent, then the resolution drops to half of Nyquist rather than the .7 of Nyquist that would pertain if the two Green subarrays are identical in response and treated as such.
Oct 11, 2009 at 10:16 AM
Previous versions of ejmartin's message #7626082 « Canon EOS 7D Master thread »
