RE: gamma 1.0 and loss of detail
RE: gamma 1.0 and loss of detail
- Subject: RE: gamma 1.0 and loss of detail
- From: "CS Carl Stawicki (4211)" <email@hidden>
- Date: Thu, 21 Feb 2002 09:24:47 -0500
Hi Steve, and everyone.
I'm sorry if this wasn't clear, I get high blood pressure when this subject
comes up. (breath in.........breath out..........breath in..........breath
out......). OK, I'm good.
My main point was to try and illustrate the fact that G1.0 is bad news for
an image, without even performing any curves or edits. If you were to read
the Lab values throughout the neutral RGB scale (8bits), you'll notice that,
in G1.0, many "L" values are skipped in the shadows. This fact proves why
you get posterization in the shadows of an image, because there are not
enough perceptual values to handle detail. The shadow detail has to be
"simplified."
The test I suggested doesn't represent an actual workflow, is just a way to
understand the math between different gamma's and Lab. Now, practically
speaking, the scanner profiles I've made and dealt with all have, what seems
to be, a gamma of ~2.x and higher. If I were to convert from my scanner
space to a working space, I've found it makes sense have a destination gamma
more similar to the source gamma.
(Thank you Chris Cox, Charles Poynton, and anyone else I got info from on
this subject)
Carl.
>
----------
>
From: Stephen Marsh
>
Sent: Thursday, February 21, 2002 7:48 AM
>
To: email@hidden
>
Subject: RE: gamma 1.0 and loss of detail
>
>
> I went through this !@#$ Gamma 1.0 crud myself. A simple experiment to
>
do
>
is
>
this:
>
>
1. Create an 8-bit RGB neutral blend from 0-0-0 to 255-255-255.
>
2. Duplicate it and assign each an RGB space with different gammas, 1.0
>
and
>
2.2
>
3. If you read the 'L' values down in the shadows, from 1.0 you get:
>
>
0 rgb = 0L
>
1 rgb = 4L
>
2 rgb = 7L
>
3 rgb = 10L
>
etc...(In other words, there're a lot of holes in the shadows.)
>
>
4. Convert each RGB file to LAB and check out the 'L' histograms. 'nuf
>
said.
>
>
Carl <
>
>
Carl, I think we all have been down a similar road to this road to see
>
what
>
the gamma affects are - but I have a question, if I may.
>
>
I can't understand steps 1 and 2 - they seem to go against what I thought
>
I
>
knew - which gives me cause for concern. <g> Hoping you can help clarify
>
your post:
>
>
What gamma space is the initial gradation being made in?
>
>
Would you not construct a LAB gradient from 0-100L and then convert this
>
to
>
1.0 and 2.2 gamma RGB, edit with say a curve and then go to LAB (or simply
>
read LAB values from RGB without the mode change)? Or perhaps construct a
>
grad in 1.00 gamma RGB, edit and then go to LAB, then constuct the same
>
file
>
again in 2.2 RGB and then edit and go to LAB?
>
>
In most cases the conversion to/from 1.00 gamma is the test, is it not so?
>
Assigning would only be applicable if the target image was constructed in
>
the same gamma as the profile being assigned, for a true result? Most
>
input
>
is 1.8 or 2.2 gamma, so a conversion from this to linear 1.00 gamma would
>
be
>
the way to go in most cases - assigning 1.00 gamma to an image which is
>
clearly not linear to begin with would not be a true test, if I understand
>
things? I understand the benefits of using the assign profile command to
>
make 'white lies' for colour correction purposes (changing luminosity or
>
colour, sometimes know as a 'false profile' technique) - but in this case
>
assigning 'random' gammas to a existing RGB image would not seem to lead
>
to
>
correct conclusions.
>
>
Assign 1.00 gamma to a 2.2 gamma corrected image and convert to output -
>
and
>
it is severely lightened.
>
>
Convert the 2.2 to 1.00 gamma and edit then convert to output - and you
>
loose shadow detail, posterization occurs and the image is a bit darker
>
than
>
it was originally.
>
>
My conclusion: All that 1.00 gamma data is in the highlights and not the
>
shadows - if you have images of polar bears in the snow or near white eggs
>
on a near white background - then using a 1.40 or lower gamma is probably
>
better than 2.2 - but for regular images, 1.00 gamma will lead to shadow
>
detail damage faster than anything else. 2.2 and 1.8 gamma working spaces
>
are so much better for the majority of images and input/output.
>
>
Sincerely,
>
>
Stephen Marsh.
_______________________________________________
colorsync-users mailing list | email@hidden
Help/Unsubscribe/Archives:
http://www.lists.apple.com/mailman/listinfo/colorsync-users
Do not post admin requests to the list. They will be ignored.