[argyllcms] Re: help with camera profile

  • From: Ben Goren <ben@xxxxxxxxxxxxxxxx>
  • To: argyllcms@xxxxxxxxxxxxx
  • Date: Wed, 25 Jul 2012 10:01:03 -0700

On 2012-07-25, at 9:37 AM, Iliah Borg wrote:

> You are editing in the output space, CIE. No, that is not how photography 
> works ;) We use density calibration to control output.

Only if you're one of those ancient fuddy-duddies still painting with charcoal 
on cave walls!

> Colour is a given, it is defined by the set of dies, pigments, colour filters 
> in front of the sensor. It is easy to determine the spectral/colour 
> properties of a CFA - all you need is a rather cheap monochromator, a light 
> sphere, and a reference photodiode ;) Or, some inside information from the 
> manufacturer. Or, some red, green, and blue patches to shoot.

Would the red, green, and blue from a ColorChecker work *well* for what you're 
describing? Barring, that, could you expand a bit on the monochromator / light 
sphere / photodiode path? Assuming, of course, that that's something within the 
realm of a mere mortal without access to a well-equipped R&D lab....

> Look at characteristic curves, they state density for the exposure, and do 
> not care for colour. Colour is the second step. The error most are making 
> when trying to imitate film in raw processing is that they do not realize the 
> curves are not taking colour into account.

I know there are those here looking to imitate film, but I'm personally doing 
this for art reproduction / giclee printing / copy work. I've been able to get 
very good results so far with the hand-massaged DNG profile + Argyll profiling 
that I described, and early indications are that I can do even better with RAW 
Developer. But if there's better yet to (reasonably) be done, I might as well 
skip straight to it.

>> What I'm envisioning for Argyll is a modification to scanin such that it 
>> still takes an RGB TIFF but then calculates the calibration curves at the 
>> same time as it reads the patch values. You could then feed it a linear TIFF 
>> output if your RAW developer supports such, or whatever else you can manage, 
>> and let scanin be the one to worry about the tone curve.
> This would be my wish list: First, equalization of light based on the target 
> border (includes white balance and light un-evenness).

Robin Myers's EquaLight does a superlative job of that, but it only works on 

(Actually, it doesn't do white balance...I've been able to get great results 
with white balance by cropping to a neutral target (PTFE) and tweaking until 
the histogram in ACR is as tall and narrow as possible. I'll check the RGB 
values of known near-neutral patches as confirmation. Not sure yet how I'll do 
it with RAW Developer.)

> Second, linearization using the grey step wedge.

Would such linearization apply equally to all channels? As I wrote, I'm getting 
good results today with the frankenstein approach, and that only uses the 
single tone curve in the DNG profile. I'm imagining, though, attempting to 
figure out the spectral characteristics of the color filters in the sensor and 
getting Home Depot to mix up some step wedges that closest match (the 
complements of) those colors.

> Third, direct coding of desired gamma or tone curve (say, tone curve based on 
> an amp file using L* from it).

..and then fourth profiling with Argyll as a final, presumably minimal 

> Nothing is particularly hard if you divide it into small jobs (c) Henry Ford

...until you wind up with so many small jobs that managing them itself becomes 
a big job....


Other related posts: