Deriving Curves from an Image (Technical)

02/13/2010 11:24:13 AM · #1

Silly title -- stick with me a moment.

I have two images. One is the original, more or less straight from the camera. The other is something I did in Lightroom (trial) and I have no idea what I did or how I did it. All I know is I love the color scheme and the way the colors were "mapped." Given two images, and software like Photoshop or PSP, is there a way to take the two images and somehow get the color mapping from the original to the edit?

For starters, I tried taking the original and the edit and channel splitting them each into three images (RGB). I took the red channel from the edit and overlaid it on the red channel from the original. Then on the edit, I chose blend mode Difference. So this gives me an idea how the reds of the image are mapped. Now what? It would be nice to reverse interpolate a red curve from the Difference channel. Then the same for green and blue...

I'll stop there (for fear of alienating the few of who've gotten this far) and ask if anyone has any ideas.

02/13/2010 11:31:17 AM · #2

Open both images in Photoshop. On the original image, dupe the BG layer then open the "match color" dialogue box. Where appropriate in that dialogue, specify the lightroom image as the source image and click "OK". That should do the trick as far as making the colors on the original match the lightroom colors.

Or bring it close anyway, there may be issues such as contrast etc that keep them from mapping exactly.

R.

sfalice

Canon PowerShot SX60 HS

02/13/2010 11:45:12 AM · #3

Originally posted by Bear_Music:

Dang, you're a genius, Bear. Great solution.

02/13/2010 12:20:30 PM · #4

Interesting -- I don't have Photoshop, so I'll have to figure out how this translates in PSP X2 or GIMP.

Also, my intention was to get a set of curves that would allow me to apply this color scheme to any image.

Sorry if I wasn't clear on those points. Thanks for looking at it, Robert...

02/13/2010 12:27:39 PM · #5

Originally posted by bvy:

Well, that's effectively exactly what "match color" does, if you're working in photoshop and you also have a curves layer you can copy over. Any adjustment layer in any photoshop image can be dragged over to any other photoshop image and applied, so...

How you deal with this in your situation I don't know, but in my case I actually DO have some stored "master images" that I use for match-color purposes. It also works very well for swiping duotones from other photographers :-)

R.

02/13/2010 12:38:21 PM · #6

Let me expand on the above/beneath/more recent post:

Let's presume you even *could* generate such a set of curves for each channel in RGB, and apply those curves to any image. The end result would STILL not be the same from image to image, because you start from a different point. The red channel info, for example, tells you how much to amp or mute the red channel, but it's dependent on what the red channel looked like in the first place. So an image with more-saturated reds will respond differently to that particular curve than an image of lower-saturation reds. Does this make sense?

Now here's another thought that occurs to me:

There is an interesting feature of Topaz Detail, where you can adjust tones in the image in the cyan/red range, the blue/yellow range, and the magenta/green range. These sliders don't change the colors, but they change the luminance of the colors, so that if you slide the one to more-yellow-than-blue, the yellows and browns get brighter and the blues get darker. And so forth on the other channels. And these can be saved as defined presets.

So if there is a Topaz detail for the program you use (I am not sure if there is), then you can set all the "detail" adjustments to zero, thus not affecting the "grittiness" of the image at all, adjust the tone sliders to what works for you, adjust the saturation to what works for you, and have a savable preset that answers some of your problems. If this were in Photoshop, then this preset plus "match color" would answer all of them.

But there may be a more elegant solution of which I am unaware...

R.

02/13/2010 01:18:45 PM · #7

Okay, I need a while to process all that.

Regarding your first point, though, wouldn't an image with more saturated reds just have higher red values? How do you define "more saturated" in RGB terms? I don't pretend to be an expert on these things, so help me understand. Seems like mapping red values from X to Y shouldn't have different results on different images providing we're starting with unedited image files.

For example, I use these curves to simulate cross-processing and get a consistent effect regardless of the image.
//www.photoshopsupport.com/tutorials/or/cross-processing.html

Olympus OM-D E-M5 Mark III

02/13/2010 01:45:14 PM · #8

Originally posted by bvy:

Regarding your first point, though, wouldn't an image with more saturated reds just have higher red values? How do you define "more saturated" in RGB terms? I don't pretend to be an expert on these things, so help me understand. Seems like mapping red values from X to Y shouldn't have different results on different images providing we're starting with unedited image files.

I'm over my head also :-) It made sense to me when I typed it, but the above has me wondering if I'm right. I'm not in a position to *check* it right now or anything, so... You're essentially suggesting that the channels are absolutes, that you can bring new absolute values to the original, and I'm thinking they are relative, that they modify what's in the original. I don't know which of us is right.

I bet Kirbic knows...

R.

I shot a PM off to Kirbic, we'll see if he weighs in.

Message edited by author 2010-02-13 13:47:55.

GinaRothfels

Olympus 8mm f/1.8 M.Zuiko Digital ED Fisheye PRO Lens

02/13/2010 02:01:23 PM · #9

I've got a program on my computer which I've hardly used called TONE HACKER. I think it does what you're looking for. I've tried a Google search, but I'm not sure which is the right link and I don't have time to go through them now. See what you can find.

Canon RF 24-70mm f/2.8 L IS

02/13/2010 02:11:14 PM · #10

Thanks, Gina. I'll check it out.

Robert, I guess we both stand to learn something. Kirbic has been summoned.

ETA: Just now seeing your edit. Guess he'll get the message!

Message edited by author 2010-02-13 14:12:47.

kirbic

Canon EOS R5

02/13/2010 02:37:48 PM · #11

Hey all,
I actually saw this thread earlier, LOL. If an image is adjusted using curves only, it should in theory be possible to reconstruct the curves adjustment given the starting and final image. The caveat is that you'd have to compare them on a pixel-to-pixel level, so if the edited image has been resized, you'd be out of luck.
I actually thought that Robert's "match color" solution was ingenious. My approach would have been to proceed like the OP did. Take the red channels, for instance, from the original and edited images. Place the one from the original image on top, set blend mode to difference. Create a curves adjustment layer, and adjust the original to match the edited (difference is all black). Record the settings for the curves layer. Lather, rinse repeat.
When done, combine the settings. Record an action to perform the adjustment, and presto change-o, you've got the desired output.
Again, this works as long as the images are the same size. If they are not, it still may work if you resize one to match the other, but you may never get "all black." You might get halos or other artifacts along edges.

kirbic

Canon EOS R5

Canon RF 24-70mm f/2.8 L IS

02/13/2010 04:05:40 PM · #12

OK, been playing a little with this, and yes, it works well. I overlooked one element in my original post. Here's a detailed breakdown of the procedure:
- Open both images (must be same size, up-sample smaller one if required)
- Double-click on the background layer on the "original" image and rename it to "Original"
- Do same with the edited image, bur rename it "Modified"
- Drag one of the layers to the other image, while holding the SHIFT key (keeps it registered to the canvas). NOTE: it does not matter which one is on top!
- Set the top layer's blend mode to "Difference"
- Select the layer named "Original" << important to make sure the "Original" layer is selected
- Select the Channels tab
- Highlight the Red Channel
- From main menu select Image>Adjustments>Curves and adjust until you have as close to a pure black image as possible
- Write down the Input and Output values for the points on your curve
- Repeat this procedure for the Green and Blue channels
Once you have a table of input and output values, you can create a curve that exactly replicates what you used to "adjust away the difference." You can create this as an action to be applied to any image instantly.

02/13/2010 05:31:58 PM · #13

That's brilliant -- I see why you were called upon. I'm going to try it out. Thanks for stepping in!