Hello everyone,
Has anyone experience about restoring faded color photographs from
positive print, using darktable?
There is high-expertise information here:
https://photo.stackexchange.com/a/22727/34215 ... which also implies lot
of work and manual adjustments.
Knowing that prints are based on subtractive color synthesis, it seems
natural to model the problem as if some of the inks were linearly (or
even non-linearly) faded. But RGB being additive, common "multiply"
operation (e.g. "channel mixer" module) doesn't fit this model. With an
"add" technically it would do, but it would be practically too awkward
to adjust both "multiply" and "add" sliders on each change (see below
for what I've found).
What does darktable provide as options?
I have tried the following darktable modules :
* Color balance (interesting how a negative "lift" answer the "multiply"
+ "add" issue mentioned above, see effect on histogram).
* Color correction (cf.
https://www.darktable.org/2012/03/color-correction/). I tried it and it
is IMHO unsuitable.
* Color zones (unsuitable IMHO).
* Color mapping.
The best result I had so far involves approximate manual steps to get to
a point where the "color mapping" module can pick up and nicely do the
bulk of the work.
Method
* In dartkable open the faded image.
* Open "levels" module, click "auto" then adjust the black point to
avoid too black shadows.
* In "color balance" perform manual approximate correction, mostly using
negative lift sometimes combined with gamma.
* The important criteria of the previous two steps are (1) get an
overall correct luminance (2) correct the strongest aspects of the color
bias, even if the result is not fully satisfying. For example, get skin
tones that mostly look like skin, sky that mostly looks blue, etc. Show
image side-by-side with a modern photograph to get a comparison point.
* The important step: follow
https://www.darktable.org/2013/04/color-mapping/ using as source a
modern photograph with good color balance that features the same
elements (in my case: blue sky, wood, skin, foliage). Worry that it
won't work. Adjust "color dominance" and suddenly get a decent (possibly
good) result.
Results
This method worked rather well, with moderate manual work and produces
results resembling the modern photograph used as color reference (which
is the point of the "color mapping" module). Obviously it's not the
intended use case for the "color mapping" module.
Interestingly, the "color dominance" parameter appears to have some
"threshold effects": from 0 to 100%, result starts awful (color casts
applied to wrong areas), does not change at all for some ranges, and
suddenly changes to something rather nice, then back to something
unacceptable and dramatically different near 100%, staying at that when
reaching 100%. Based on https://www.darktable.org/2013/04/color-mapping/
I would expect monotonous change and the more relevant results happening
with low values. Perhaps this behavior comes from a too approximate use
of the "color balance" module and would theoretically disappear if that
step was done "perfectly"?
Limits
The limits of the approach are that it has some "apply whatever color"
(from modern photograph) rather than "restore original image colors"
flavor.
Also, many faded photos have areas less faded. Most of them near the
edges, sometimes spots anywhere in the picture. These remain visible on
end result as blobs of strong color change.
Others directions to explore?
Yes, doing once, applying on many.
It may be interesting to use the original vs processed image to
calibrate a simple (non-linear but monotone) model of ink fade. Results
may be expressed as parameters for a "tone curve" module, or even as 3
instances of "base curve" operating on R G and B (as shown fully
manually on https://www.youtube.com/watch?v=RQXbmC_v2d4 , you should
probably mute the sound). The resulting module parameters could be
applied to other photos from the same batch and get decent rather good
color correction without more work. Obviously it would get as biased as
the color difference between the photograph used for calibration and the
modern source photograph used as color reference, which may or not be a
problem, depending on the goal.
In the meantime, I could get some correct results by performing the
method on one photo and copy-pasting the "color mapping" module on other
photos with mostly the same subject. On others photos, it still kind of
works on some images. At least it can be tried to instantly get
something on n photos at the cost of one copy-paste, letting me
disabling it only on those photographs where it is not satisfying, which
is still a benefit from working on all the photos manually.
Has anyone also tried to restore color from old scanned faded paper photos?
Any comment welcome on any part of this endeavor. Thanks!
--
Stéphane Gourichon
____________________________________________________________________________
darktable user mailing list
to unsubscribe send a mail to [email protected]
