Wow - blows my mind

Jacob Rus's new curve-based technique, with examples
mikemeister_admin
Posts: 4927
Joined: Fri Sep 20, 2013 8:29 pm

Postby mikemeister_admin » Tue Sep 02, 2008 12:28 pm

There is an interesting posting

http://tech.groups.yahoo.com/group/colortheory/message/20162

that blows my mind and gives a lot of control.

I tried it on the London Regent St and it was really quite neat.

So can anybody deduce what on earth is happening!

Enjoy
Chris

ggroess
Posts: 5342
Joined: Wed May 24, 2006 2:15 am
Contact:

Postby ggroess » Tue Sep 02, 2008 3:36 pm

so this is a ploy to make everyone join the yahoo group...

LOL

Greg

-default
Posts: 1916
Joined: Thu Mar 26, 2015 1:53 am

Postby -default » Tue Sep 02, 2008 5:45 pm

Very interesting.  I haven't seen anything like this before - gotta try this later today.

I've often considered having an expanded axis for Lab corrections, since so many of the corrections, for a and b particularly, involve just a small range of curve values.

mikemeister_admin
Posts: 4927
Joined: Fri Sep 20, 2013 8:29 pm

Postby mikemeister_admin » Tue Sep 02, 2008 6:25 pm

I've played with it a little and tried it on the London photo and changing the colour of the car - both with great success.

It seems, to my mind, to be a super masking technique, so by selecting the channel one wants to duplicate across all channels, one effectively restricts the changes (colour or tone) to that mask.  What I find interesting and I didnt know before was that the Linear Light will act as a sort of Soft Light control for curving (in tone or colour).  It certainly is odd to be curving in the vertical direction to alter the image and one begins to want a line ability, rather than forumulated gentle curves, when points are altered - in my experiments I was for ever having to tie most of the curve back to the horizontal.

It seems a super way of doing things in odd situations, but I can not see that it would be very useful in the 'normal' run of pp-ing, but I need to experiment more.

I would love to hear your comments when you have a chance to try it out.

Regarding your comments about the a/b curving, then I have often thought that a 'zoom' control would be useful to enable one to alter curves more gently - not sure that it will give us the ablility to be able to do anything differently, but it would certainly be a great MMI feature.

Tother thing that I frequently use, and why this posting interested me, is the BlendIF facility.  BlendIFs adds that bit more control over which areas are changed.  However this idea blows BlendIFs completely out of the water.  What my mind can not grasp is how I can combine the power of CM with it's masks + this idea to have almost complete control over my corrections.  I really must re-examine the Linear Light blend mode (which I had dimissed as a gimic).  These strange (?) blend modes seem to becoming to the fore, what with Dan's idea of Vivid Light blend.

I think it is marvelous, that people are finding more and more ways of playing with corrections, when I had assumed that most of them had been discovered by now.


mikemeister_admin
Posts: 4927
Joined: Fri Sep 20, 2013 8:29 pm

Postby mikemeister_admin » Tue Sep 02, 2008 6:51 pm

and, of course, the other really useful feature that CM would bring to this, is that ability to interactively curve the choosen channel - super power!

mikemeister_admin
Posts: 4927
Joined: Fri Sep 20, 2013 8:29 pm

Postby mikemeister_admin » Tue Sep 02, 2008 6:55 pm

damn it - how can I use CM as a curving tool, as a grouped adjustment layer, on the 3 channel layer?

-default
Posts: 1916
Joined: Thu Mar 26, 2015 1:53 am

Postby -default » Tue Sep 02, 2008 10:30 pm

My guess is as a smart filter.  I'll let you know when I've tried it out.

ggroess
Posts: 5342
Joined: Wed May 24, 2006 2:15 am
Contact:

Postby ggroess » Wed Sep 03, 2008 1:54 am

Mike,
Can you move this topic to another area so it does not get lost when the boards are cleaned out..

Also more people need to see this that are not on the class board.



Greg

derekfountain
Posts: 251
Joined: Fri Jan 26, 2007 1:24 pm

Postby derekfountain » Wed Sep 03, 2008 7:16 am


There is an interesting posting

http://tech.groups.yahoo.com/group/colortheory/message/20162

that blows my mind and gives a lot of control.


Could someone post a copy of the content being discussed? The colortheory board is a closed one, so most of us can't see the post Chris is talking about. The alternative route - subscribe, read one post, unsubscribe - seems rather a lot of hassle for a lot of people!

mikemeister_admin
Posts: 4927
Joined: Fri Sep 20, 2013 8:29 pm

Postby mikemeister_admin » Wed Sep 03, 2008 7:22 am

Ah sorry, I thought it was open to all to read.

First post
>>>>>>>>>

Hi List,

This has possibly been discussed before (I haven't run any
exhaustive searches of the list archives, and haven't more than
skimmed Dan's book in the bookstore, so let me know if I'm saying
something well known) but I haven't ever seen it mentioned in
other books or in the various color-correction internet material
I've seen, so here goes.

For problematic images--i.e. those with differing color casts in
highlights, midtones, and shadows, or possibly casts appearing in
only particular hues--or for artistic effect, the blend if
sliders (or alternately various sorts of masks) are useful for
limiting the range of various curves corrections, etc. But I
have stopped going to them first for complex corrections in these
cases, in favor of the following, which is sort of my way of
hacking Photoshop into a color correction GUI I find more
intuitive than using the Curves tool "straight":

* Convert the image to CIELAB (obviously)

* Copy the L* channel (I

* Create a new layer, and paste the L* channel into all three
channels (L*, a*, and b*).

* Set the blend mode of this new layer to "Linear Light,"[†]
and bear with me here, because at this point the image will
look like a black/white posterized splotch. I usually name
this layer something like "L*-based corrections".

* Add a new curves adjustment layer, and set it as a clipping
mask.

* Make all three curves--L*, a*, and b*--completely flat, i.e.
Input 0 -> Output 50, In 100 -> Out 50 in the L* channel,
and In -128 -> Out 0, In 127 -> Out 0 in a* and b*. At this
point you should have an image that looks exactly like the
original, because Linear Light blend mode makes no change to
the bottom layer where the blend layer is middle gray.

* Now here's the fun part: make some points along those curves
and slide them up and down. You can thus adjust (linearly
shift) L*, a*, and b* values for various background levels of
L*, with much finer control than using the blend if sliders,
or any other tool I know.

- Want more shadow contrast? Leave the point at i0->o0, and
add another couple at, e.g., i6->o53 and i25->o54.
- Shadows a bit too green? Make an upward bump in the left
end of the a* curve.
- Sky not blue enough? Make a downward bump in the right
end of the b* curve.

&c.

That's it. If you want to make corrections based on the a* or b*
values, instead of based on L*, the procedure is identical, except
copy the a* channel (or whatever) into each one of the new layer's
channels.

The main down-side to this approach for me is that Photoshop's
Curves controls here have a very big range, and the corrections
to be made with this tool tend to be *very* close to that center
line: I sometimes wish I could zoom the curve's vertical axis to
only show its middle third (or in the case of adjusting based on
underlying a* and b*, zoom both axis like that).

Anyway, I realize this email was a bit long, but hopefully others
find this tool as useful as I have. It becomes especially nice
after setting up an action to create these layers, but making
them manually isn't *too* arduous.

Cheers,
Jacob Rus

[†]: A few other blending modes also work, but have nearly the
same possibilities here, and linear light is easier to use
in my experience. But feel free to experiment :-)

P.S. This is my first posting to this list, so I should probably
introduce myself. I'm a Harvard undergraduate government
concentrator--formerly in the math department, but I got tired
of problem sets, took a year and a half off to work on various
programming stuff, and am now back for the last year.5 of school.
I have been a huge Photoshop fan for about half my life now, and
I try to get out to take pictures as often as I can with my D50
(never often enough). I continually tell myself I need to put
some up online, but I somehow manage to never get around to
registering a domain and coding up the simple site I want.
Sometime soon though, I promise!

P.P.S. Can anyone recommend the best introductory/intermediate
book about image processing? I just this week started trying
to learn how to use NumPy (array math for Python) so I can try
to build some of my own types of tools, mostly for own
edification, but also because I sometimes find Photoshop's color
correction tools limiting in ways that seem easy enough to
improve on, but I want to "put my code where my mouth is" so to
speak. :-)


SECOND POST
>>>>>>>>>>

Okay, a few follow-ups (which are left here because I forgot to
mention a couple of them, a couple are more detailed than needed
for the initial message and would possibly impede comprehension,
and I just figured one out right now).

1. This approach (copying one channel into all of the channels,
setting blend mode to Linear Light, and then adjusting a
flat curve) can also be used in RGB and CMYK to good effect.
In those color modes, I basically use it as a more powerful
channel mixer + blend if combined. If you've been
unsatisfied with the power available in channel mixer, I
recommend you play with this, for example for adding back
detail to very colorful objects.

2. This is a way to mix L*/C/M/Y/K channels into RGB images,
and vice-versa. Just duplicate the image, convert to the
new space, copy the desired channel, paste it back into all
channels of a new layer set to Linear Light, add the
clipping adjustment layer, and adjust to heart's content.

3. For something even potentially cooler, try:

- Duplicate the image
- Convert the duplicate to CIELAB
- Use the Hue/Saturation tool to rotate the hue 30°[†]
- Convert the image back to RGB or CMYK
- Copy one of the resulting channels
- Go back to original image
- Paste into all three channels of a Linear Light layer
- &c.

This lets you mix channels which are half-way between RGB
and CMY (roughly: modulo the differing luminosity
contributions of different wavelengths), which means you
can alter colors along sometimes more convenient axes than
RGBCMY.

4. Various channel mixing uses of this technique can be very
effective for converting a color image to grayscale.
Indeed, the combination of this technique, the
Shadow/Highlight tool, and various sharpening/blurring
of various channels on large/small scale, is enough to
handle pretty much all my B&W conversion needs. :)

5. Once you have a B&W image, the technique can also be an
effective tool for tinting it with extreme control. Just
(in CIELAB mode) set up a layer with L* copied to all
three channels (no need to set this to Linear Light this
time), and then create a curves adjustment layer, with
the L* curve left as a normal ramp from i0->o0 to
i100->o100, and then set the initial curves of the a* and
b* channels to be flat 0 all the way across. Then move
a* and b* curves up and down to tint various L* levels
of the image different tints, without affecting the L*
component of the resulting colors.

6. I said before:

> The main down-side to this approach for me is that
> Photoshop's Curves controls here have a very big range,
> and the corrections to be made with this tool tend to
> be *very* close to that center line: I sometimes wish
> I could zoom the curve's vertical axis to only show its
> middle third (or in the case of adjusting based on
> underlying a* and b*, zoom both axis like that).

I just figured out how this can be solved. Woohoo! We can
make a Levels adjustment layer between the 3-alike-channels
layer, and the nearly-flat-curves-meat-of-the-technique
layer, and use it to "zoom" the horizontal axis, by bringing
in the input left/right sliders by the desired factor (if
we're trying to edit based on a* and b* values, they are
often very low, so we can bring compress the levels here to
1/5 or even 1/10 of their original extent. This lets us use
more of the curve in our main curves layer.

Likewise, to "zoom" on the vertical axis, we can add a Levels
adjustment layer *above* our main curves layer (note all of
these layers should be set as clipping masks), and bring its
*output* sliders (on both ends of every channel) in to 1/5 or
whatever their original extent, which will effectively dampen
whatever adjustments we make in our main curves layer, which
means we can see the adjustments we're making better, because
they won't all be so close to the axis.

I'm excited about this. It makes using this technique much
faster because less precise mousing is needed.

Okay, enough for now. I'll give others a chance to weigh in (e.g.
to tell me they think I'm completely nuts). :-)

Cheers,
Jacob Rus






Return to “Jacob's Ladder”

Who is online

Users browsing this forum: No registered users and 3 guests