[UFRaw-Devel] Banding on the Canon 7D

[Robert K. <rl...@al...>]

Reference this discussion:
http://forums.dpreview.com/forums/readflat.asp?forum=1019&thread=36942985&page=1

There is an issue with vertical banding being observed on the Canon
7D, typically when underexposed RAW images are pushed by several stops
(either selectively in the shadows or the entire frame).  This issue
is *not* observed when JPEGs from the camera are similarly adjusted,
or when DPP (Canon's imaging software, including RAW converter) is
used.  It has been reported with Adobe Camera Raw, and I've reproduced
it using ufraw, which is using dcraw 1.439 (9.05).  There is some
evidence that this isn't happening on the 60D or the 550D/Rebel T2i,
which are newer but otherwise use the same sensor, but it's not
entirely clear from the discussion.  There is also some discussion
that the pattern of the banding is the same from frame to frame.

This makes me suspect that it's some kind of column-to-column
variation in the low bits.  This would not be perceptible under normal
circumstances, but if the low bits are stretched by pushing the
exposure, they become visible.  Since this doesn't happen with JPEG
files or when images are processed with DPP, it suggests that there's
information in the RAW file (perhaps in the black frame pixels) that
can be used to undo the banding.

I have some RAW files that show this; one was exposed at -2 stops (for
bracketing purposes) at ISO 100, and with +4 or +6 stops of exposure
compensation in ufraw, the banding is very apparent.  The other was
exposed normally at ISO 200 (in a somewhat abnormal setting) and it's
possible to see banding at +1 stop.

I've placed one of the files in ftp://66.92.65.9/pub/img_2399.cr2
(warning: 22 MB, and I only have 384 Kb/s upload speed, so it will
take a while to download).  Look at the shadows at the bottom of the
frame with exposure compensation of at least +4 stops.  Or, let me
know if there are other tests you'd like me to run.