[argyllcms] Re: Profile input white not mapping to output white

  • From: Gerhard Fuernkranz <nospam456@xxxxxx>
  • To: argyllcms@xxxxxxxxxxxxx
  • Date: Fri, 23 Nov 2012 23:40:11 +0100

Hi Ben,

seems you are facing an extrapolation issue beyond the media white point. The 
.ti3 file covers only the RGB 0...53 range, and there are no data points in the 
53...100 range. Particularly non-parametric models (i.e. CLUT profiles) can 
suffer severely when they need to extrapolate such a big range (-aG would 
likely do a better extrapolation job here).

Furthermore the CIE and RGB data in the .ti3 don't fit well to each other. If 
this .ti3 file was created from 706C8559.CR2, then I would not be surprised, as 
this shot seems to be not the best one (uneven illumination - about 1 f-stop 
difference between upper right and lower left corner of the target).

Best Regards,

Am 23.11.2012 20:05, schrieb Ben Goren:
On 2012-11-23, at 9:50 AM, Ben Goren wrote:

I just looked a bit closer, and I can only see a half-dozen pixels clipped in 
the second green channel in patch N18, and they're all outside of the 
BOX_SHRINK margins.
I've done a bit more experimentation, and I'm now *VERY* confident that this 
has nothing at all to do with the RAW processing and everything to do with 
something going wonky in colprof.

I've created plain flat TIFFs in various ways out of the original RAW file, 
confirmed that there's nothing weird in the TIFFs (no clipping of anything but 
the black patch, etc.), and then fed those TIFFs to Argyll, and used Argyll for 
all further manipulation. I've also used Photoshop to tag the image with the 
profile and gotten the same results.

Here's a more dramatic recent example:

That's the original file tagged with the profile. You can confirm by checking 
the RGB values of the background -- they're all 255,255,255, but they get 
turned cyan by the profile. Assign any standard generic-ish profile (such as 
sRGB) to that image and the cyan becomes white.

For reference, here's an idealized synthetically-created version of the chart. 
You can confirm at a glance that the chart itself is being accurately 

Indeed, if you do a blink comparator type of analysis between those two 
pictures, there's hardly any visible difference at all.

Here's the .ti3 file, generated from a full-sized (but cropped) 16-bit TIFF of 
the (untagged) top image:

Here're the two most relevant lines from that file:

  #     XYZ_X        XYZ_Y       XYZ_Z     RGB_R    RGB_G    RGB_B  STDDEV_R 
I13  8.7720e-03  8.5310e-03  8.8570e-03  0.52812  0.49257  0.47701  0.19195  
0.17432  0.14688
N18 94.683      98.339      81.005      53.085   53.987   52.269    6.8720   
6.6649   6.7166

I13 is the light trap I described earlier, and N18 is a bit of Teflon thread 
seal tape. The latter is only barely above 50%, with a looooooong way to go 
before clipping.

And here's the profile by itself, created by colprof without any options:



Other related posts: