Computer Graphics with Photoshop - Strange curves effect

Author

Strange curves effect - bug or feature?

Peter Wollenberg

2005-08-19, 4:14 am

I was playing with curves a little bit (CS2) and got a strange effect
when I applied an admittedly rather uncommon curves transformation to
a monochrome gradient. You can see the result at:

http://tinyurl.com/8xh98

the image displayed links to the actual PS file with the gradient
layer still in place, which you may download for inspection. The
bottom left part shows how the image looked like at the beginning and
the top right what I got after adding my contrived curves layer. You
will find that the gradient is made up between two colors with exactly
the same hue and saturation, and with brightness varying between zero
and 100%. Since I did not modify individual color channels I'd have
expected a result which is actually obtained when the curves layer
blending mode is set to luminosity. This is what I do routinely to
prevent unwanted changes in saturation.
What I got here is really surprising to me, since there is not only a
change in saturation as expected but also in hue. It looks to me as if
there were some serious rounding or overflow error in the curves
routine.
BTW, I opened the file in CS and 7 as well and it looked always the
same.
Please share your opinion

Peter

Mike Russell

2005-08-19, 7:14 am

"Peter Wollenberg" <jenelisepasceci@web.de> wrote in message
news:4305801e.2572203@134.96.4.2...
>I was playing with curves a little bit (CS2) and got a strange effect
> when I applied an admittedly rather uncommon curves transformation to
> a monochrome gradient. You can see the result at:
>
> http://tinyurl.com/8xh98

Although the gradient is monochrome, it does contain color - just the same
color. Your image is a great example of how the RGB master channel alters
hue. Try the same curve on the lightness curve of Lab mode and you'll see
no hue change at all.
--

Mike Russell
www.curvemeister.com

Warren Sarle

2005-08-19, 7:14 pm

In article <4305801e.2572203@134.96.4.2>,
jenelisepasceci@web.de (Peter Wollenberg) writes:
> ... Since I did not modify individual color channels

Ah, but you did. The RGB curve is applied to each channel
individually, after applying the separate R, G, and B curves.

--

Warren S. Sarle SAS Institute Inc. The opinions expressed here
saswss@unx.sas.com SAS Campus Drive are mine and not necessarily
(919) 677-8000 Cary, NC 27513, USA those of SAS Institute.

Peter Wollenberg

2005-08-22, 7:15 pm

"Mike Russell" <RE-MOVEmike@Curvemeister.comRE-MOVE> wrote:

>"Peter Wollenberg" <jenelisepasceci@web.de> wrote in message
>news:4305801e.2572203@134.96.4.2...
>
>Although the gradient is monochrome, it does contain color - just the same
>color. Your image is a great example of how the RGB master channel alters
>hue. Try the same curve on the lightness curve of Lab mode and you'll see
>no hue change at all.

BTDT ;)
Well, the lesson I learned is to use curves adjustment layers
exclusively and set the blending mode to luminosity in order to retain
hue and brightness. I still don't understand the mathematics behind
the whole thing completely. I'd have anticipated the complementary hue
(H-180 deg) to appear but actually I got the whole spectrum. And I
can't see that this behaviour of the curves dialog is good for
anything. The dialog might well offer the luminosity instead of RGB as
the master curve, because most people don't want or even expect
changes in hue, but rather want to lighten or darken parts of the
image without a color change when applying curves.
Thanks to you and Warren

Peter

Mike Russell

2005-08-22, 11:14 pm

"Peter Wollenberg" <jenelisepasceci@web.de> wrote in message
news:430dca7d.20384968@134.96.4.2...
> "Mike Russell" <RE-MOVEmike@Curvemeister.comRE-MOVE> wrote:
>
>
> BTDT ;)
> Well, the lesson I learned is to use curves adjustment layers
> exclusively and set the blending mode to luminosity in order to retain
> hue and brightness. I still don't understand the mathematics behind
> the whole thing completely. I'd have anticipated the complementary hue
> (H-180 deg) to appear but actually I got the whole spectrum. And I
> can't see that this behaviour of the curves dialog is good for
> anything. The dialog might well offer the luminosity instead of RGB as
> the master curve, because most people don't want or even expect
> changes in hue, but rather want to lighten or darken parts of the
> image without a color change when applying curves.
> Thanks to you and Warren

Although I favor using the Lab color space, luminosity is an excellent way
to achieve the effect. For minor adjustments I still find myself using the
RGB curve with no bad effects. Certain underexposed images also fare well
with the RGB master curve.

Interestingly, luminosity still screws up for CMYK images.