Alexey,
From what you're saying your results seem to be different from mine, but you
also did some things differently.
I didn't try to fine-tune colprof settings, but thanks for pointing me in that
direction. I was building a matrix-only profile, which, as I understand, is the
best choice with the low number of patches (24). In your example the profile is
LUT based, which could explain the much lower error magnitude you're getting
for the patch colors. I'm not sure how well this result extrapolates to other
colors though.
When you're saying the images look close, what exactly do you mean? Could you
measure the L* value of the middle gray patch please?
Cheers,
Nic Nilov.
On Apr 29, 2022, at 14:04, Alexey Gribunin <Gribunin@xxxxxxx> wrote:
Hello Nic!
I tried your datas and got quite good results.
Here are my strings:
scanin -v -F 2622,1399,6880,1425,6880,4257,2582,4227 -p -dipn -G 1.0
chart-raw.tiff CC24Classic.cht CC24Classic-D50.cie
colprof -v -qh chart-raw
I've used your big file directly (usually I'm downsampling such big files to
2-3 megapixels).
I've got quite good profile:
Profile errors: max. = 2.817343, avg. = 0.686372, RMS = 0.942445
After that I've made color conversion from RGB to LAB:
cctiff -v -ia chart-raw.icm chart-raw.tiff chart-raw_LAB.tif
Also I've embedded profile into RGB TIFF.
cctiff -v -ia -e chart-raw.icm chart-raw.tiff chart-raw_RGB_Embedded.tif
Both LAB-TIFF and RGB-TIFF files look very close in Photoshop.
They looks completely close after RGB-TIFF file conversion from RGB to LAB in
Photoshop.
So everything seems working correctly.
Best regards,
Alexey Gribunin.
On 28/04/2022 21:36, Nic Nilov wrote:
Hello,
I've been trying to create a simple camera calibration profile and more or
less succeeded (profcheck reports errors max. = 5.159180, avg. = 2.285364,
RMS = 2.613159).
I'm now trying to apply this profile to an image of the CC24 target (the
same the profile is derived from) and am observing some inconsistencies I'd
like to find an explanation for.
When I do it via this cctiff command:
cctiff -v -ia chart-xyz.icc chart-xyz.tiff chart-xyz_corrected.tiffthe result is looking very nice. Notably, the middle gray patch measures L*
50% with near zero color deviation. The resulting tiff has CIELab encoding,
and opens as such in Photoshop, so it needs a conversion to an RGB
colorspace before further processing. This concerns me as LAB has some
encoding issues which may be relevant even with 16bit images. I would prefer
the cctiff output to be in an RGB colorspace.
So, when I specify the -e option to embed the ICC profile like this:
cctiff -v -ia -e chart-xyz.icc chart-xyz.tiff chart-xyz_corrected.tiffexactly that happens. The output tiff now has RGB encoding with the profile
embedded.
I open this file in Photoshop using the embedded profile, at which point the
image looks significantly darker, the middle gray patch measures L* 35.
I then convert the image to ProPhotoRGB colorspace using the Absolute
Colorimetric intent (which is the same as cctiff -ia does), but the image is
still darker than the CIELab version. The middle gray patch now measures L*
45.
If, instead of using cctiff, I open the input chart-xyz.tiff file in
Photoshop directly, assign the same ICC profile and then convert to
ProPhotoRGB with Abs.Col., I get exactly the same looking image as with the
previous step (cctiff -e).
The main question is why when the image is in an RGB colorspace the look is
darker?
The second one is does the use of CIELab as the output encoding present an
image quality concern (primarily due to quantization with the resulting
banding and gamut reduction) when the first step of the post-processing in
the Photoshop has to be the Lab to RGB conversion?
Thank you,
Nic Nilov.
