Christian Macey writes:

>> Again see above but yes it's easy to go to RGB from LAB, but one does
have to be careful with out of gamut colours. (I think I need Stephen
Marsh to bail me out on this one. But at the present time I think he's
a bit busy looking after Dan's list.) <g> <<


Thanks for the vote of confidence Christian. <g>

Yes there is degradation to a move to Photoshop ICC LAB with 8 bpc or higher
data - plus quantization and rounding of pixel values on the roundtrip back
to the original mode. Starting with a greyscale 2.2 gamma image with a full
range of all 256 levels, moving to LAB will render 236 unique available
levels (http://www.brucelindbloom.com/LevelsCalculator.html). An interesting
side note, the LAB quantization damage (invisible to the naked eye) looks
similar to JPEG quantization when differenced/equalized.

Since this is a lossy move at the data level, is this a visually lossless
move?

In almost all most cases - yes.

I have only come over one case where there was a noticeable visual change to
the file from the simple RGB > LAB > RGB trip with no LAB edits, this was a
blue sky which changed saturation/brightness somewhat - enough to be
considered an artifact or an error. I have not been able to reproduce this
issue and do not have access to the image (this goes back a few years).

Do not expect RGB value perfect conversions between RGB and LAB (rounding
issues aside) - it depends on the flavour of RGB and it is possible to have
RGB values which are not possible in Photoshop's ICC LAB mode (as opposed to
CIE LAB, which is even worse to edit in than ICC LAB from a user
perspective). Most have come to expect no visual or other problems
associated with the LAB move, if a file was going to posterize or band or
whatever there is it most likely would have without the LAB trip.

I think of the process as being similar to applying max quality JPEG 2000
lossy compression to an image, one tries to avoid it where possible but if
one has to do it - the sky will not fall in.

If one checks the Apple ColorSync Users archives, one will find thousands (I
kid you not) of posts on why LAB mode is the Antichrist and why it should be
avoided at all costs, although there is less issue with high bit data for
these users. Most of the users on the CSU list do not see a valid reason for
using LAB when there are good RGB editing spaces available. There is some
good info here. The other extreme is on the Applied Colour Theory list,
where Dan Margulis will use LAB for some very clever moves - as he cares for
visual results more so than for intangibles such as unique levels counts and
other statistical analysis. So it often comes down to the use and workflow,
as well as how the user feels about certain issues (visual results or data
concerns).

Both viewpoints are valid, I like to keep a middle road approach -
attempting to understand the pitfalls and benefits of LAB - while using LAB
or HSB type principles while in RGB or CMYK and leaving the really tough
cases to LAB (color, luminosity, hue, saturation blend modes).

For sharpening, I would recommend that most RGB images be left in RGB and
the sharpening be performed on a luminosity blended layer or with luminance
fades directly after the filtering operation, also with layer blending
option sliders (the luminance blend is similar to LAB in most important
respects). LAB does treat darker tones a bit differently than some RGB
spaces, so for some subject matter or sections of an image, the LAB sharpen
may be more pleasing than a similar luminance blended sharpen in RGB. For
CMYK it is often best not to change modes for a sharpen, as black plate and
specific ink build data will be lost (CMYK has K, which can be even better
for an extra hit of USM or curvework etc).

LAB, CMYK and RGB are all a bit different in how they treat blending and
edge aliasing. I can foresee some potential benefit in resizing in LAB mode,
although similar (perhaps better, perhaps worse) effects could perhaps be
gained from resizing in a different gamma space or with advanced colour
settings gamma1 blending. Perhaps not all images would benefit from this
treatment, but some may.

I think that reducing pixels is just as important as enlarging pixel
counts - but upsizing seems to get all the press.

More links to LAB topics can be found here:

http://members.ozemail.com.au/~binaryfx/PSTV_links.html#L


Regards,

Stephen Marsh.


===============================================================
GO TO http://www.prodig.org for ~ GUIDELINES ~ un/SUBSCRIBING ~ ITEMS for SALE

Reply via email to