[argyllcms] Quality of camera profile - colour banding

  • From: Milan Knížek <knizek.confy@xxxxxxxx>
  • To: argyllcms@xxxxxxxxxxxxx
  • Date: Sun, 27 Jul 2008 11:52:36 +0200

Hi all!

referring to my previous post about some troubles with camera ICC
profile based on IT8 target:

The trouble is colour banding at skin tones. It appeared just on very
few images, while majority of the photo shooting day was just fine. This
banding is not visible with other profiles.

On this web page [1] there is an overview of cropped and re-sized (50 %)
image processed with various ICC profiles in UFRaw:

1. Home made ICC profile from IT8.7 target produced by Wolf Faust [2] -
four years old, used occasionally and stored in dark and dry place.
Target was captured at sun light. Before using argyllcms, the image was
manually white-balanced and gamma of 0.45 (with linearity 0.10 in
shadows) applied in UFRaw. The target is a bit damaged in few dark
patches, I have corrected these glitches in CinePaint with clone-tool. 
Parameters for argyll's profile command were -qu -as, peak error 36,
average error 2.8. I also tried CLUT instead of matrix profile, but no
big difference.

In addition to the gamma correction, the sample image has a base curve
applied before conversion to sRGB to boost the mid-tones a bit. UFRaw
applies the base curve to all channels equally. There is a visible
colour banding on the lady's shoulder where lightness changes from light
to shadow.

(The photo was shot shortly after storm rain. The lady stands under pine
trees, which make a mixed lighting conditions with the cloudy sky in the
open area in the background. Depending on white balance, the colour
banding is either more or less visible.)

2. Same as 1, but to increase the effect of colour banding, wavelet
denoising (in UFRaw) of 200 was applied. The banding is more visible
now.

3. This time, the profile for Canon EOS 10D produced by Ture Palsson
[3]. It is a linear profile, so gamma and linearity was set to zero, but
I have still applied a base curve to boost mid-tones. This image does
not show apparent colour banding as in 1 and 2.

4. The profile is taken from Canon's Digital Photo Professional raw
converter bunch of profiles with method described here [4] - I used the
Neutral one. The colours are different, which is an expected behaviour,
since DPP does some further undocumented adjustments during conversion.
However, the shoulder does also not show the colour banding.


Obviously, I would like to be able to create a camera profile without
this banding - any recommendations?

Thanks in advance and best regards,
Milan


[1] http://www.milan-knizek.net/files/tmp/banding.html  The profile and
log file is available for download, too.
[2] http://www.targets.coloraid.de/
[3] http://www.lysator.liu.se/~ture/eos10d/
[4] http://maashoek.nl/color_profiles.html


Milan Knizek
knizek (dot) confy (at) volny (dot) cz
http://www.milan-knizek.net - about linux and photography

Other related posts: