PDA

View Full Version : Here's a good color-correction Photoshop tip



chris jordan
4-Aug-2006, 15:43
Hi guys, I thought I'd pass along a useful LAB color-correction technique that I use, that not many people know about. It's something that can't be done in RGB mode, at least without about a 15-step masking process that is not as accurate as a simple curves adjustment in LAB.

The technique is to reduce the saturation of low-saturation colors, without affecting the high-saturation colors. In RGB this would be called a "saturation curve," which is a tool that does not exist in Photoshop.

For example, in a photograph of a city, you want grey concrete things to be grey, not tinted with blue or purple or cyan or whatever, and you want your shadows to be neutral instead of blue or purple or red. But, you also want the blue things to be blue, and the cyan things to be cyan, etc. Using the standard RGB color tools, it can be difficult to remove the color from these neutrals while maintaining correct color balance and saturation overall.

Being able to do this is especially important for users of transparency films, because of tranny films' high sensitivity to color. Tranny films will almost always put too much color in the neutrals and blacks, which can give the image an over-saturated look. If you reduce saturation overall in RGB mode, that will take color out of the things that are supposed to be saturated, making the whole image look soft and dead. The technique I'm talking about here will allow you to take the color out of the low-saturation greys while leaving the colored objects untouched.

It's a simple curves adjustment in LAB, done only in the two color channels. Here it is: First go to the "a" channel, which controls the green-magenta spectrum. Lock down the center point (0,0), and then lock down a series of points up and down the line from 10 upwards, and -10 downwards. So you should have a point at 10, 10, then one at 14, 14, and one at 20, 20, and one at 24, 24. Then do the same thing with the negative values, placing points at -10,-10, -14,-14, -20,-20, etc.

Now place a point on -4, -4, and change the "output" number to 0, so the point is now at -4,0. Place another point on -8, -8, and change that output number to 0. That will have the effect of taking everything in the image that is slightly green (green values of 0 to 8), and taking all the green out, without affecting the things that are very green (with green values of 10 or more, which are locked down).

Now do the same thing to the positive side of the curve, which is the magenta side. Place a point on 4,4, and then change the output number to 0, and do it with 8,8 also. That will have the effect of taking everything in the image that is slightly magenta (with magenta values of 0 to 8) and neutralizing those to zero, without affecting any of the saturated magentas that have magenta values of 10 or more.

Now anything in the image that had a slight green or magenta cast is perfectly neutral grey, while all the saturated greens and magentas are untouched. Because of the way that LAB works, this tweak will apply to the whole tonal scale at once-- it will take a tiny green or magenta cast out of the highlights, midtones and shadows simultaneously; everywhere in the image that has a slight amount of green or magenta will be neutralized to a color value of zero.

Then you can do the exact same thing on the blue-yellow continuum on the "b" channel. The "b" channel adjustment is particularly useful because most tranny films throw too much blue into the neutrals and shadows. This tweak will fix that without taking any blue out of the things in the image that are supposed to be blue. It works especially well for images that have a lot of neutral greys, such as fog, cement, grey overcast skies, etc. Interestingly, after applying this curve, the colors frequently will appear more saturated, because you have neutralized the neutrals and now the colored objects will contrast more with the clean neutrals and the whole image will sparkle.

And of course you don't have to do all the colors at once; you can correct only the blues this way if you want, or only the magentas or whatever. Just lock down the other side of the curve and work on the side you want to neutralize.

If you do it as an adjustment layer you can fade it back to taste, and of course you also can adjust the curve to taste-- if 8,0 takes out too much color, then you can try 8,4 or so, which will be a softer version of this adjustment. You will have to flatten the image before converting back to RGB or the adjustment will be lost.

I usually do this as a final step to make sure the neutrals are nice and clean before resizing, sharpening and printing.

Cheers from Seattle,

~cj

www.chrisjordan.com

Ron Marshall
4-Aug-2006, 16:16
Thanks Chris, that will save me much time in the future.

I've played around a bit with LAB since reading "Canyon Conundrum", it is a powerful colorspace, but I have lots to learn.

bob carnie
5-Aug-2006, 06:12
very nice tip,
Do you find and increase or decrease of contrast in these open shade areas with the colour sucked out of them?

Leonard Evens
5-Aug-2006, 07:15
I just tried the basic idea with the Gimp, which I use under Linux. (I will try it under Phtoshop when I boot into XP.) I work almost entirely with negative film, so I don't usually have oversaturation problems, but it is sometimes difficult to remove slight color tinges from neutral areas. The method does seem to work, but I do have a question. In reality, there is seldom a true neutral color in a scene. There are just elements that one expects to be neutral. For example, concrete on buildings can be just a trifle yellow while asphalt can be slightly blue or some other cool color. My problem is that the color tinge may vary over a surface and that doesn't look right. It is a challenge maintaining the subtle differences in different shades of grays and whites while removing any variations in tint on any single surface. Your trick may help by desaturating all colors close to being neutral, but it would have to be applied very carefully. Alternately, I could apply the technique to selected regions separately.

Any ideas about this?

julian
5-Aug-2006, 08:35
Any ideas about this?

Another way of doing this is to use the iCorrect Editlab pro pluggin, which works in lab and is an easier workaround and should do what you want.

chris jordan
5-Aug-2006, 09:29
Bob, one of the nice things about LAB is that color and density are controlled on separate channels, so when you do something to the color it has no effect on image contast or brightness. WHen working in RGB mode you can duplicate that particular effect by putting your layers in "color" mode instead of "normal" mode. Every layer I make in Photoshop is always either in "lunimosity" mode (which affects the brightness but not the color), or "color" mode (which affects the color but not the brightness).

However, the saturation curve I talked about above can not be duplicated in RGB.

Leonard, this issue doesn't come up as much when working with neg film-- I have found that neutral tones in negative film tend to be much more accurate, and I usually like the fine variations in color that appear in the neutrals. Negative film will show a slightly yellow concrete with a slightly blue shadow, which can look really beautiful, so when working with neg film I generally don't do this LAB correction. And it always has to be used carefully; otherwise subtle tones get lost even working with tranny film. But it is a good way to soften oversaturated neutrals and make an image look more natural.

QT Luong
5-Aug-2006, 13:07
That's a great tip Chris, and it answers the question of why you needed to go LAB space.

paul stimac
5-Aug-2006, 21:50
I'm having a tough time getting the 4,4 and 8,8 (-4,-4 and -8,-8) points in. When I click inbetween 0 and 10 the 10 moves. How do I avoid this?

chris jordan
6-Aug-2006, 00:01
You can click some points higher up on the scale, and then move them down by entering the numbers manually (key in the actual numbers), or by using the arrow keys to get the points down to where you want them.

Once you get the curve set up, you also can hit the "save" key and save that curve so you can call it back up again. It's enough of a pain to construct that I did that awhile back-- just saved the curve as "Lab Neutrals" and then click on "load" every time and it pops right up.

bob carnie
6-Aug-2006, 07:32
Chris

Funny enough I have two interns from a French photography school on my premise using our Lambda.
One of the fellows is a big fan of yours., in fact he is shooting large format negative with lots of concrete and funny enough , colour casts in the nuetrals.
I printed your first post, scratched out and indications on who posted and gave it to them.
I wanted to see if they got it. immediately they dissected the notes and were applying the method you described in LAB and applying it to their test prints for final prints.
It is amazing how the world has flattened and a post in Seattle is being used three hours later in Toronto by two very young and talented individuals.
Thanks for the tip.

Now I will get the two kids to teach me how to use it. this will only take me two years.

chris jordan
6-Aug-2006, 10:14
Bob thanks for checking in with that great story. The internet is just amazing that way-- it has connected us in ways that we couldn't have dreamed of just a few years ago. Please send your two young friends my warm regards.

~cj

DrPablo
6-Aug-2006, 10:44
I'm a big fan of LAB for color corrections, and I've read Dan Margulis' great book on the subject.

I do have a big concern with this technique though.

For those of you unfamiliar with the details of LAB, 0 is neutral in the color channels (A and B). Negative values in A are green, positive values magenta, and in B negative values are blue and positive values yellow.

The technique described above takes every color value in LAB between -10 and +10 and neutralizes it. This is a BIG deal, because you're taking literally 10,000 color values (10 x 10 x 10 x 10 for the pos and neg values in each color channel) and rendering them neutral. In fact it's more than 10,000, because you also have the 256 luminosity values (in 8-bit), so you're actually taking 2.56 million colors and neutralizing them.

The following image shows colors that all have A and B values that are between -10 and 10. The luminosity values are different, but they're irrelevant. Every single one of these colors, whether from a color cast or not, would become gray with this tecnique.

http://www.pbase.com/image/64703373.jpg

Furthermore, there are many colors that have a value of under 10 (or -10) in one channel but greater than 10 in the other channel. Like a yellow-green or a purple. If you neutralize the value that's under 10 but not the one that's greater than 10, you will get a color shift.

Now, I do use LAB for color casts, but I use a different technique -- I measure neutral objects for any color bias and kill that specific cast. Again, neutral objects (white or gray) should have values of 0 (or very near 0) in both the A and B channels if they're neutral.

After measuring them, simply take a single point on the A and B curves and drag it up or down until you've removed any color cast. So if your white object has an A value of -6 (making it green), drag -6 up to 0 (you can plug in the numbers too). This will kill the green cast.

Jack Flesher
6-Aug-2006, 10:52
In RGB this would be called a "saturation curve," which is a tool that does not exist in Photoshop.
www.chrisjordan.com

Hi Chris:

You outline a nifty technique and this is in no way intended as a flame, but in addition to the concerns brought out by Dr. Pablo above, I wanted to address the comment you made above because it is not accurate.

In RGB mode if you add a curves adjustment layer above your image and set the blend mode of that curves layer to "saturation," you in fact have exactly a "saturation curve" tool.

If you flatten that curve on all channels to a full horizontal line, all outputs at 0 (or all outputs the same), you will get a completely grayscale image. Alternatively, you can of course also set points for any single hue and pull them down to any desired degree of saturation.

Cheers,

chris jordan
6-Aug-2006, 10:53
Dr. Pablo, I think you might have misread my original posting. My technique is to lock down the 10,10 values, so that they are unaltered. Only the values lower than that get neutralized-- the 2,2's, 4,4's, etc. Usually I don't neutrallize the 8's all the way-- they get taken from 8 down to 3 or 4 or so. And then, after that, I fade the curve back to taste, which is usually 50% or so. I talked about all of that in my post; hopefully it was more clear to others...

chris jordan
6-Aug-2006, 10:54
And one other thing: My technique is not for removing color casts-- that is easy to do in any of several different ways in RGB mode. The point of the technique I am describing is to reduce the saturation of the neutrals, which cannot be done in RGB mode.

DrPablo
6-Aug-2006, 11:04
Ah yes, I did misread the lockdown at 10. This would be better, but there are still some subtle colors that would be neutralized (all these have A and B values from -4 to 4).

http://www.pbase.com/image/64704622.jpg

But you certainly could make the colors more subtle, like reprogramming -2 to -1 and -4 to -2, etc, making the color curve shallower without truly neutral. This would preserve some color information. Of course it's to some degree subject specific.

Jack Flesher
6-Aug-2006, 11:05
The point of the technique I am describing is to reduce the saturation of the neutrals, which cannot be done in RGB mode.

Chris: It can be done in RGB and quite easily. Please read my post directly above -- you probably missed it while responding to Dr. Pablo.

Cheers,

chris jordan
6-Aug-2006, 11:15
Hi Jack, yeh that technique kind of works, but not the same. You can change the saturation of any particular hue, but you can't divide out the saturated colors from the unsaturated colors, and adjust them separately along a curve, which is what you can do in LAB. There is no way in RGB to desaturate the neutrals while leaving the saturated colors untouched.

The other thing is, all I'm offering here is another tool for the tool box, to be used however it helps. It doesn't work with every image, and it may not be useful for some people at all. It works for me, though, most frequently with images made on transparencies in low light, where the neutrals are too saturated but otherwise the image looks properly color balanced.

Jack Flesher
6-Aug-2006, 11:22
but you can't divide out the saturated colors from the unsaturated colors, and adjust them separately along a curve, which is what you can do in LAB.

Actually you can by building the curve much the same way you build your Lab curve, locking down points you don't want to change -- plus it is easier to load a mask at the same time in this process allowing even more control over the areas you are affecting.

Again, I am only pointing this out for clarity and am in no way challenging the veracity of your Lab method -- both are indeed good tools to know and which tool one chooses to use will be dependant upon their normal image processing workflow.

Cheers,

chris jordan
6-Aug-2006, 11:30
Jack I can't get it to work in RGB-- where do you place your points? In LAB, the neutrals are all right around zero regardless of their brightness; but in RGB the neutrals are all up and down the curve depending on their brightness. So how do you lock down the saturated colors and tweak just the neutrals?

chris jordan
6-Aug-2006, 12:21
Jack, by the way-- your book is going out in tomorrow's mail; should be there Wed or Thurs.

cheers,

~cj

DrPablo
6-Aug-2006, 12:26
You could do it in RGB, but it would be a pain and sloppier.

In an individual color channel (red, green, blue) the more saturated colors are given higher tonal values than less saturated colors. So a bright blue sky will be very light in the blue channel and dark in the red channel. You can use the grayscale channels as masks to further desaturate the near-neutral areas. Problem is you'd have to either work individually with each channel, or you'd have to make some kind of composite.

I don't think the advanced blending options would work too well -- they'd be better in LAB.

Jack Flesher
6-Aug-2006, 12:44
Okay, I understand you now Chris -- and your method is better at targeting the neutrals specifically.

However it's not really tough to generate the mask I mentioned.

1) Open your image
2) Go to Select>Color Range
3) Click on your main neutral area
4) Now activate the dropper+ tool and click on all additional neutral hues you want to desaturate.
5) once you have them all click okay
6) add the curves adjustment layer and the mask is automatically loaded based on your selections
7) change the blend mode to saturation and pull down the top right corner of the curve to taste.

And thanks for getting my book out so quick -- I am looking forward to it!

paul stimac
6-Aug-2006, 15:43
dang it... I still cant get those values. (0,0)(4,0)(8,0)(12,12)(16,16)(20,20)(24,24) *(and their negatives) are as close as I can get them. It just won't let me do the (10,10). This seems to give the effect your talking about, but its frustrating that I can't enter the correct values.

chris jordan
6-Aug-2006, 16:37
Paul, that should do it; as long as the upper numbers are locked down, and you have the ones closer to zero moved down to zero, that's it. 8,0 is a bit harsh; I'd say maybe go 4,0, and 8,3 or something like. Then make sure 12,12 is locked down, and above that is all locked down, and you'll have the tweak I'm talking about.

Create the curve as an adjustment layer, and fade it back to taste (I usually fade it back to somewhere in the 50-75% range, so I'm not taking all of the color out of the neutrals).