[argyllcms] Re: Profiling between two images

  • From: Pascal de Bruijn <pmjdebruijn@xxxxxxxxx>
  • To: argyllcms@xxxxxxxxxxxxx
  • Date: Wed, 4 Feb 2009 08:35:49 +0100

On Tue, Feb 3, 2009 at 9:47 PM, Gerhard Fuernkranz <nospam456@xxxxxx> wrote:
> Pascal de Bruijn wrote:
>> The gamma doesn't cut it... The shaper is a quite important part of the 
>> profiles
>> in question.
> Yes, I see, obviously the camera applies an S-shaped TRC.
>> I never expected exact results, with regards to my camera (or DPP) 
>> processing.
>> I have attached a .ti3 file, which gives completely different shaper
>> when executed
>> like:
>>   colprof -qu -aS
>> Than when it's executed like:
>>   colprof -qh -aS
>>   colprof -qm -aS
>>   colprof -ql -aS
>> In all four cases the matrix seems nearly identical.
> I did take a look at your data now.
> Colprof -qh -aS bails out with "Powell failed" (no profile was
> generated). Obviously the optimization did not converge... [You did not
> encounter this problem, did you?]

I didn't have this problem, at all...

> The options -qm and -qu resulted in shapers which do not differ very
> much (see the green and blue curves in the attached diagram). They look
> pretty reasonable up to the target's Dmin (about RGB=230...235).
> However, the extrapolation beyond Dmin up to RGB=255 (i.e. the very
> steep ascent) does not look reasonable at all. I'm also not sure whether
> the profile might possibly clip some shadows. I would actually not call
> these -qm and -qu shapers bumpy, IMO they are reasonably smooth.
> Option -ql gives indeed a different (less reasonable) shape. I guess the
> -ql curve [which have fewer degrees of freedom than the -q m/h/u curves]
> is not "flexible" enough to fit the given data, but on the other hand I
> find it more bumpy than the -qm -and -qu curves (which is rather a
> contradiction).
> Btw, I was using V1.0.3, just in case that the version makes a difference.

I'm using 1.0.3 as well...

Another question, how did you genereate the graph?

I manually inspected the profiles using LProf... your method seems a lot
more convenient.

Pascal de Bruijn

Other related posts: