On 06/10/2014 22:27, Eric Stevens wrote:
On Sun, 05 Oct 2014 22:37:46 -0400, Alan Browne
wrote:

On 2014.10.05, 20:55 , PeterN wrote:
On 10/5/2014 6:57 PM, Alan Browne wrote:
On 2014.10.05, 14:42 , PeterN wrote:

We went through all this some many months ago. I demonstrated clearly
that the amount of 'loss' was negligible in practical terms.


I would use the terem "color change." anstead of loss.

Any change is a quality loss. Whether that is colour difference, tone,
brightness, sharpness ... whatever, it's a loss.


Then you are using a different definition of quality.

Not at all. A non lossy process would have:

RGB-A -- X-format -- RGB-B

with RGB-A identical to RGB-B

But - the fact is that with Lab

RGB-A -- Lab -- RGB-B

RGB-A =/= RGB-B, therefore there was quality loss.


But hang on: we do accept a certain degree of quality loss as part of
the normal process of editing. It doesn't take much manipulation to
turn a smooth histogram into something like
http://pe-images.s3.amazonaws.com/ba.../fix-white.gif
Push things a bit harder and you can get
http://www.snoopy.me.uk/misc/365proj...gram/comb3.jpg
or even https://aperture64.files.wordpress.c...09/combing.gif

The production of histograms like the first one is common and
generally acceptable. The second histogram is worse but even then may
be acceptable. Only the last one is so bad that it will nearly always
be unacceptable. The point of all this is that some degree of quality
loss is virtually inevitable as soon as you start to manipulate an
image.

In the context of the present discussion, the question is, does the
conversion to Lab colour incur any more damage than one can expect in
the course of ordinary editing? My understanding of nospam's claim is
that it does. My (admittedly limited) experience with it suggests that
conversion to Lab causes no significant damage; certainly less than I
am going to inflict on the image by the changes I want to make.

Since CIELAB is a colour space intended to manage just noticeably colour
differences more optimally than the naive RGB colour space it isn't too
surprising that you cannot *see* a difference in the final JPG taken
from RGB or via CIELAB. But they are very slightly different.

As to the extent of the damage, I can only refer to my original
experiment described in Message-ID:


------------------------------------------

This one continues to bother me. I am still inclined to agree with Dan
Margulis. I'm not quite sure what procedure Andrew Rodney is proposing
to prove his point so, using Photoshop CC, I have carried out my own
test as follows:

1. Find a JPG with a suitable range of colors. This one came from my
wife's collection:
https://dl.dropboxusercontent.com/u/...20IMG_2154.jpg
I saved a copy as a PSD (see below for the reason).

2. Copy and convert to Lab. I couldn't save to JPG from Lab so I saved
to PSD. See
https://dl.dropboxusercontent.com/u/...54-via-Lab.jpg

3. I then loaded the two PSD files into a new file as separate layers.
(1) above was the background layer and (2) was the next. I subtracted
the 2nd layer from the first with the result shown in
https://dl.dropboxusercontent.com/u/...Difference.jpg
That's right: solid black.

It is only solid black so long as you don't use the histogram tool to
look in detail at the noise floor (also I am not sure how you did the
differencing - you may be missing half the differences if you did a
simple subtraction which clipped to zero as opposed to an absolute
difference where any discrepancy is rendered as a positive difference).

Use the histogram tool and you will see that the images do differ in the
luminance least significant bit. This would be undetectable in practice
but it is non-the-less a difference (ie not lossless).


4. To confirm the point I took a screen shot. See
https://dl.dropboxusercontent.com/u/...t%20Screen.jpg
Note the histogram. All of the pixels appear to be down at the zero
end of the scale: that is, jet black.

The only conclusion I can reach is that there is no difference between
a PSD created from a RGB file and a PSD created from the same image
when it has first been converted from RGB to Lab.

In practice you will not be able to see the difference and without pixel
peeping you can't see the difference on a simple difference image but it
is still there - just below your visual threshold.

I'm not wedded to the perfection of the method I have used and I would
be interested to hear from anyone who has a meaningful criticism.

The only thing that did surprise me was that the resulting errors are
entirely in luminance there is no chroma noise introduced at all.
(this might be an artefact of how you did the differencing)

--
Regards,
Martin Brown