As far as I know NX still uses ICC profiles. Maybe they still can be
extracted using this technique:
These ICC profiles expect that the raw file has been preprocessed in
some way (applying some gamma etc), RawTherapee has this coded in
rawimagesource.cc, but it was a while ago since it was tested last I
think, but you get the idea.
So what I would do to replicate Nikon NX look is to extract those ICC
profiles, and figure out how the preprocessing is made. Then I'd
generate a new ICC or DNG profile which makes the same translations but
without the preprocessing (will require some coding).
Note that ICC-using raw converters often do some sort of preprocessing,
Capture One does it as well. This means that to make a ICC profile for
a specific raw converter you need to know what preprocessing it does.
If you do a normal profiling workflow you get the preprocessing baked
in so it's no problem then, but if you create an ICC from scratch you
need to apply the preprocessing in that process.
On 07/27/2015 02:35:40 PM, Adriaan van Os wrote:
Graeme Gill wrote:
Adriaan van Os wrote:
You could try using a real photographic test chart like theColorCheckerSG,
but this has it challenges, and would limit the gamut and detailpossible
in characterizing the two RAW processors.
My thinking is, not to use a test chart, but the existing photo
library that a photograher has on
his hard-disk and let statistics do its work to compensate for lack
full test-gamut in single
photos. So, we have a JPG with a known profile and we can convert the
JPG RGB pixel values to XYZ.
And we have a raw-converter produced TIFF with different colors. So,
the task is to calculate the
color transform that would produce the same XYZ vales for the TIFF as
in the JPG. That color
transform then constitutes the Profile to be assigned to the TIFF
after the raw converter has done
its work. This is repeated over many photos and a statistical average
(mean or median) is
calculated. The standard deviation would be a meaningful value too.
How would an algorithm that does this, roughly look like, apart from
the statistics, so for a
single photo combination ? I assume the algorithm is much like what
color matching systems do, or
with reverse math.
Adriaan van Os