[argyllcms] Re: noise and gamut, input profiles with B2A table

Klaus Karcher wrote:
I'm just trying to gain experience with tiffgamut, profile -g and icclink -G and tried to test the impact of noise on the gamut size / gamut mapping.
Did you gain experience with the effects of downsampling, denoising or bluring on typical image gamuts? Did you thing about adding some kind of smoothing or denoising options to tiffgamut? 

No, I've not tried playing with such things, although I've thought
about downsampling as simply a way of speeding up creating a gamut
from a large collection of images. It has a fairly dramatic
effect on your image, but I think there is no right or wrong here.
If you take a gamut of a downsampled image, than you are making
the judgment that clipping the gamut of fine detail is OK,
and that it's better to do this and minimize the impact of gamut
mapping on lower frequency components. The effect on the overall
gamut mapping may be very subjective and image dependent, I don't
known, and it would need some work to figure out what is reasonable.

Of course if there really is a lot of noise in the image, then
this may well be the right path. But isn't that something that has to be
decided on a case by case basis ?

This moves down the path of spacially adaptive gamut mapping, something
that's been played with a few times in the literature. It is an interesting
area, and
Besides, I came across a little issue with the argyll scanner profile I used: Photoshop's denoise filter doesn't work with it as the profile has no B2A table. This is not a big issue: as a workaround one can temporary attach a matrix profile or convert the image to the working space, but sometimes I prefer working with the image in its native colorspace without workarounds or redundant conversions. 

It's not required by the ICC spec., and would take extra time to generate,
so I left it out. I was also debating as to the correct way to make scanner
profiles. Technically I think I should do the opposite of what I'm
currently doing: I should be making a rspl of the CIE -> device values,
then inverting it to create the device -> CIE. I'm currently cheating
(as I suspect almost all other profile software does) and simply
applying the rspl to the device->CIE pairs.
IMHO the option to add B2A tables or additional matrix tags to input profiles would be nice feature. Is this possible without to much hassle? 

It's not trivial, but easier than changing things for printer profiles.

Graeme Gill.

Other related posts: