[argyllcms] Re: help with camera profile

  • From: Stephen T <stwebvanuatu@xxxxxxxxxxxx>
  • To: "argyllcms@xxxxxxxxxxxxx" <argyllcms@xxxxxxxxxxxxx>
  • Date: Thu, 2 Aug 2012 07:07:22 -0700 (PDT)

"Cannot the problems with shadows/highlights be related with the dynamic range 
of the camera?"

No single exposure of a test chart can fully exercise the dynamic range and 
colour gamut of a modern digital camera. What we need is a profile model (and 
workflow) that can extrapolate beyond the bounds of the test chart and 
consistently deliver pleasing results.

Linear transforms are well-behaved and predictable. I read somewhere 
that Adobe DNG profiles are essentially matrix profiles (with a LUT for 
tweaking if needed) - someone please educate us if I am wrong.
I have not yet experienced shadow problems with matrix profiles. My profiles 
render shadows too light and I simply apply an RGB tone curve (in the RGB 
working space) to darken the shadows, increase contrast in the mid-tones, 
smoothly roll-off the highlights and even adjust exposure.

Highlight clipping (L* > 100) during conversion from camera RGB to the working 
space RGB is more difficult. One workaround is to underexpose the raw data (in 
software) and then correct the exposure and compress the highlights in the RGB 
working space. Adobe's DNG profiles seem to apply this strategy. To check this:
1) Take a correctly exposed photo of a test chart.
2) Apply a DNG profile without any exposure correction or tonal adjustments.
3) Check if L* = 50 for any middle grey patch. I find L* << 50 (in Raw 

With linear exposure scaling and linear matrix transforms, the above process is 
no less accurate for test chart patches. In real world photos, highlights and 
deep shadow  colours are based on extrapolation.

For general photography, matrix profiles seem to perform well, meaning results 
are not far from the CIE standard observer (with appropriate exposure 
adjustment if required, see above). For well-controlled studio work, where the 
illuminant is fixed and highlights are controlled, then LUT-based profiles 
should be more accurate but perhaps more limited?


Other related posts: