ejmartin wrote: ejmartin wrote:
Did a little more analysis of UCSB\'s colorchecker RAW file. Here is the fractional difference (G2-G1)/(G2+G1), after median filtering both channels to reduce the column gain fluctuations:
The deviation from pure gray is an indication of channel imbalance. The patches that seem to be different are the black patches, where we are dividing by a small number which amplifies fluctuations; but the rest is remarkably uniform, spot checking indicates the fractional difference of the two greens is a few tenths of a percent, which would not be enough to cause serious nastiness.
This is quite different from skibum\'s first copy, which showed a substantial difference in yellow in one direction, and a difference in cyan in the opposite direction. I\'ll have to go back and dig up my data on that one, but I seem to recollect the difference was several percent. Here, UCSB\'s copy doesn\'t show any dramatic difference between yellow and cyan patches.
Tentative conclusion would be a bad batch among the first run of cameras which afflicted skibum\'s copy but not UCSB\'s.
skibum5 wrote:
what do you make of even his copy showing 8-20% greater StdDev for the Green2 channel than the Green1 channel though? On mine I notice 15-20% while on all my other canon bodies it was mostly about 3% difference or so.
I dunno. Is the difference in std dev correlated with anything? For instance is it greater when there is a larger difference of G-R and/or G-B? I seem to recall someone having a theory that the column readout swinging back and forth between color channels could affect the noise.
I have to go back and check to see if it varies per shade or not (it might just vary per total green signal in a direct way though).
Basically if I use Rawanalyze to select a block and then take the avg value and std dev of only green 1 for the patch or only green 2 for the patch the differences for both are larger with the 7Ds than with all the other canon bodies I have ever used. With the 7Ds it\'s like 0.5-2% for the avg values and 10-20% for the std dev while for the other cameras (none of which maze easily it\'s more like 0.05-0.5% and 2-5%). Where the average is higher the StdDev is always lower. All the 7D seem to have higher StdDev for G2 than G1 (not random body to body where some have G1>G2 and some G2<G1 or G2=G1).
i\'d have to go back and check to say more though.
well it happens with the 7D and doesn\'t happen with any RAW from any other Canon body I have ever used and the ones where it does not happen, all the rest, don\'t maze to any significant degree and
Nov 12, 2009 at 10:27 PM
Previous versions of skibum5's message #7769022 « Canon EOS 7D Master thread »
