LAB: I'm a convert

Andy

One of our SmugMug pros, Thomas Manchester wrote in with a customer complaint about the print of this file:



Thomas' customer had this to say "The color on some of the girls is quite grey, especially on their legs, making them look bruised. (also on their hair)."

This image is fascinating to me on many levels. One - the skin tones are generally VERY good - and the faces lit nicely. But according to Thomas, it was a "really quick" shot, and only with one flash onboard - so I applaud the photographer for great execution - and I give a Bronx Cheer to the camera sensor for failing to make sense of the data in the deep shadows. That discussion is for another part of our forum :wink

Let's look closer at some of the problem areas, and you can see that there's some data in these areas but it's choppy and not good:



So, thanks to Rutt, and his constant hammering and yammering on LAB :lol3 - when I first started to attack this image for repair purposes, I found that my normal tricks in RGB world, while they were working, they were somewhat inefficient. I rang up Rutt and said: "teach me to fish." Thank goodness for Rutt, because we all know reading is hard - and so I asked Rutt to give me the readers' digest version on how best to repair this image.

1. Convert to LAB mode
2. Duplicate layer - set to "color" blend mode
3. Focus on a problem area - use eyedropper to select "good color"
4. Use soft-edge brush (at 100% opacity) and paint over the "bad areas"
5. Lather, rinse, repeat.
6. Advanced levels: You can use layer masks to selectively "undo" areas you don't want painted over. You can use "layer blending options" and use the sliders to remove the adjustment from areas you don't want it.



And after only about 10 minutes of work, I get this:



The big benefit of working with LAB is that we're only messing with the color, not the luminosity, not the contrast, nothing else about the image changes - and so, you can make very drastic changes, and dramatic repairs, without destroying the integrity of the image. I also think that it's totally reasonable for even busy event pros to learn this technique - it takes maybe 10 or 15 minutes to learn, and then when you need it, you have it in your bag of tricks. Don't fear the LAB!

Here is the Original File.. ok you LAB-sperts - let's see what other tricks you have up your sleeve for images such as this! :ear

LAB: It's what's for dinner :food

rutt

This was really good, Andy. I didn't know you had it in you.

For others who want to learn more, the basic trick here comes from Chapters 8 and 11 of the Dan Margulis' book, Photoshop LAB Color.

Personally, I'm really really bad with brushes and so I'm willing to work extra hard to make it easy not to make a mess when I do have to use them. When I was a kid, I always made a mess out of coloring books, and was really jealous of those who could stay inside the lines. So I gave this shot a little thought. Hmm, I'd like a mask that would help me limit the brushstrokes to the places that need them. How to make that quickly?

Here's the idea I had. It's a variation on the mask making technique from Chapter 9. I started with a copy of the B channel which shows nice variation between the skin, hair and surrounding areas. Then I used a variant of the CMD-CLICK formula from Chapter 12. The result was a mask, nice and sharp around the edges to help me keep my coloring where it belongs. The mask only took a minute to make after I thought of the trick. There were some places where it needed a little manual fine tuning with the brushes, but it was a great quick easy starting place.

Here's the mask I made (with red circles around the spots where it helped be keep my coloring inside the lines):



Here are the steps I used to make this mask:

1. Copy the B channel
2. Select the B channel copy
3. Autolevels
4. Curves
5. Use CMD-CLICK to identify some spots in the area I wanted to color and some spots outside the area. Type 0 into the output field of the areas I want to color and 100 into the output field of the areas I don't want to color. I end up with an inverted V shaped curve like this:
   
6. A good idea to blur, but I didn't. Say Gaussian blur, radius 3.
7. Now I can use for a layer mask for coloring.

Tom K.

A wonderful technique. I used it on the photo you posted and it worked flawlessly. Thank you very much!!

tmanchester

Thanks Andy
That was my photo, and I appreciate the work. Personally, I didn't think the image was that bad to begin with, but Andy made some nice subtle corrections to it, and I learned something in the process.

Nice job of customer service!

edgework

Hey Rutt, that's a nice mask. I need to work on getting better masks from the a and b channels.

Check out this approach. I started playing around after reading your post. Your move with the lightness curve, emphasizing the desired range and excluding the rest reminded me of how I've been making shadow, midtone and highlight masks for a while now. It suddenly occurred to me that there might be an alternate trick here somewhere. This is about 10 minutes old, so I haven't thought it all through.

First, the masks: I put a gradient map adjustment layer as the topmost layer. For a shadow mask, set the left anchor of the gradient to white, the right anchor to black. This, by definition, makes all the dark areas white and excludes the rest. For highlights, the default black to white gradient, the grayscale version of the image, gives the same result in highlights. For midtones, the fun part, set both anchors to black and set a middle anchor to white. Now you can target a range with a fair degree of precision by both positioning the white anchor to maximize the desired range and clipping unwanted areas with the right and left anchors. I shifted the white anchor around until the areas of really gray skin and hair both showed white. Now select all and copy merged with Shift-Cmd-C and paste the result into a new channel. (remember to kill the gradient map layer).

This gives a mask that covers a bit more range than the one you produced, but, being LAB, the blending sliders were pretty effective at restricting things in the desired range, coupled with a curve to pull things towards red. Given that the neutral areas were the ones being targeted, I eliminated the brighter ranges of red and yellow and most of the blues and greens. I fine-tuned with opacity and used a color layer with the same mask to help things along. The result was about the same as you and Andy came up with. Not necessarily an improvement, though the mask technique is fast and reliable. The is the first time I tried to further target a curve by adding a midtone mask with blending sliders, going after the neutrals, and, as usual, it wouldn't be possible in RGB or CMYK.

Here's the midtone mask:



and the slider settings:

Andy

You guys are great. Rutt, thanks again for the push this morning! Funnily enough, I've used this technique before, but it wasn't filed properly in my brain. So much to know!

Steve Cavigliano

While I agree that Rutt's constant promoting of Lab is a little like listening to a Born Againer....lol He sure knows his stuff and I have learned so much about using Lab from his info posted here

This issue is just another example of what you can do more easily, with less damage, using Lab that is way more difficult to do using RGB

Good work Andy and Rutt and I feel a stickey coming on. IMHO, this is a post that can help lots of people

Steve

rutt

Steve Cavigliano wrote:

While I agree that Rutt's constant promoting of Lab is a little like listening to a Born Againer....lol He sure knows his stuff and I have learned so much about using Lab from his info posted here

Thanks, Steve! Really, it isn't so much being born again as being allowed in the candy shop. I hope it will wear off eventually and I'll get excited about something else new. That's how I learn.

Steve Cavigliano wrote:

Good work Andy and Rutt and I feel a stickey coming on. IMHO, this is a post that can help lots of people
Steve

There is a pretty good thread on this topic having to do with fixing blown highlights. I realized this afternoon that this particular problem really is almost exactly like fixing blown highlights or plugged shadows except the damage is annoyingly located in the midtones where the blend-if sliders can't isolate them as easily.

Blown highlights are very very common, while I think there was something kind of unusual going on here. Andy suspects a sensor defect. I wonder if it's the camera's jpeg firmware? Or it might really be a lighting issue. The flash and the ambient light do have potential for interesting mixed casts. Combined with some sort of auto white level thing it could do bad stuff.

DavidTO

rutt wrote:

Andy suspects a sensor defect.

Andy's wrong. It's mixed lighting with two different color temperatures. What you're seeing is the shadow of the warmer source, leaving the weird parts of the picture in the shade, or cooler color temp light.

Andy

DavidTO wrote:

Andy's wrong. It's mixed lighting with two different color temperatures. What you're seeing is the shadow of the warmer source, leaving the weird parts of the picture in the shade, or cooler color temp light.

I never said "defect" I did say, limitation. BUT let's keep that discussion for the cameras forum please. You could be right about the mixed lighting - for sure - but that's not the cause of 0 or very little color data in those areas.

rutt

Anyway, I think David is right about what's going on here. And mixed casts are very common. So this might have pretty wide appeal, if not quite as much as fixing blown highlights.

For most mixed casts, figuring out how to make masks for the A and/or B channels and then using curves is the method of choice in LAB. This neutral cast, though, that's pretty annoying. Takes some serious blending to get a mask good enough to use with curves.

jfriend

Does this need LAB?

Andy wrote:

The big benefit of working with LAB is that we're only messing with the color, not the luminosity, not the contrast, nothing else about the image changes - and so, you can make very drastic changes, and dramatic repairs, without destroying the integrity of the image. I also think that it's totally reasonable for even busy event pros to learn this technique - it takes maybe 10 or 15 minutes to learn, and then when you need it, you have it in your bag of tricks. Don't fear the LAB!

LAB: It's what's for dinner

Good technique. Andy and Rutt, one thing I'm not clear on is why this needs to be done in LAB mode. There's a color blend mode in RGB. If the main part of the fix is painting on a layer in color blend mode, perhaps paired with a mask, can't all that be done in RGB mode? As I presume you and Rutt know, I'm a LAB convert, but does this fix need LAB? Clearly, when you get into some of the more advanced blend-if scenarios using the A and B channels, you can only do that in LAB, but does your first explanation require LAB or can you just use a layer with the color blend mode in RGB too?

Keep in mind, I'm also hot off the trail of the Margulis LAB, chapter 14 writeup which teaches you how to separate color from contrast in RGB mode too :.

rutt

jfriend wrote:

Good technique. Andy and Rutt, one thing I'm not clear on is why this needs to be done in LAB mode. There's a color blend mode in RGB. If the main part of the fix is painting on a layer in color blend mode, perhaps paired with a mask, can't all that be done in RGB mode? As I presume you and Rutt know, I'm a LAB convert, but does this fix need LAB? Clearly, when you get into some of the more advanced blend-if scenarios using the A and B channels, you can only do that in LAB, but does your first explanation require LAB or can you just use a layer with the color blend mode in RGB too?

Keep in mind, I'm also hot off the trail of the Margulis LAB, chapter 14 writeup which teaches you how to separate color from contrast in RGB mode too :.

You can do this in RGB, but lack of impossible colors makes the blend work less well. No real pain in doing it in LAB and you get that advantage PLUS the better blend-if sliders and the mask making tools capabilities I used.

jfriend

Makes sense

rutt wrote:

You can do this in RGB, but lack of impossible colors makes the blend work less well. No real pain in doing it in LAB and you get that advantage PLUS the better blend-if sliders and the mask making tools capabilities I used.

Makes sense. I just wondered because I was using the color blend mode for this kind of color patching before I even knew about LAB mode.

Andy

jfriend wrote:

Makes sense. I just wondered because I was using the color blend mode for this kind of color patching before I even know about LAB mode.

That's good. I was wondering if Rutt lied to me. Again. Gotta watch that guy

rutt

rutt wrote:

You can do this in RGB, but lack of impossible colors makes the blend work less well. No real pain in doing it in LAB and you get that advantage PLUS the better blend-if sliders and the mask making tools capabilities I used.

Not to mention Edgeworks mask making idea.