Correcting those newborn shots

rutt · November 28, 2005

If you haven't already, see this thread: http://www.dgrin.com/showthread.php?t=22943

Andy wants to keep the focus on time-of-shooting techniques and that's fine, but I made a quick attempt to fix in post with plate blending, etc. My results weren't very good. I wonder if someone can do a good job and provide a step-by-step?

edgework · November 28, 2005

This just might be a worst case scenario. Bad color, bad balance in the luminosity channel (way too dark) and lots of jpeg artifacts waiting to become obvious the minute you try to bring things into a manageable range.

I found no quick fix. An image like this requires surgery of the extreme makeover variety.

I initially tried using the red plate for blending into the other two, but there's just not enough detail there. It has no blotching but the shape is flat.

So I tried CMYK. The cyan plate is worthless and magenta and yellow look like different cases of leprosy, but VOILA, the black plate is not bad.

I used channel mixer on both Magenta and Yellow, reducing both from +100 to +60 and increasing black to +40 in each. Then I drew a soft mask around the face to restrict the mix.

Then I merged and went into lab. There I used a global luminosity curve to lighten everything overall, and a masked curve to enhance the color in the face (symmetrical: anchors ten units each towards the center.)

At this point the skin imperfections were demanding attention. I used the healing brush, 6px radius, and blended as much as possible.

After this it was back to CMYK for one of my favorite moves. First, a curve layer in Color mode, pulling everything down in the highlights. Then, a curve in luminosity mode. Contrast curves in the magenta channel, in luminosity mode, almost always do good things to skin tones. This one contains two points: one with Input: 80, Output 69, and the other with input 15, output 2. After that I did some final blending with the healing brush, and that's it. I was afraid to sharpen. Those artifacts are subdued, but they're not gone. It would take a lot more effort to give this baby the nice smooth skin she/he deserves. Also there is a lot of left-over grunge on the lips and probably elsewhere. But it's an improvement, however imperfect.

rutt · November 28, 2005

You got a lot farther than I did, but I may have discovered a move you missed.

I overlay blended the A channel into the L at 150% to get this:

Sttill a long way to go, but it's directly addressing the blotches. Some LAB curves to move a little toward yellow and away from magenta:

Better, but looks too orange to me now.

I also played with surface blur to address issues in her skin. This also eeems to be a good idea.

So I ended up here. Still way too dark and I can't figure out how to fix that without either making it too flat or very harsh.

This is good and hard. Maybe I can get Dan M. to look at it.

edgework · November 28, 2005

I tried a similar approach ( I blended everything I could find into everything else, at one point or another, looking for a way out of this mess ).

The problem with focusing only on luminosity is the initial whacky variation in the red tones from the blotched areas. Bringing the values into a similar range just highlights the weirdness in the cheek. It just gets worse as it gets lighter. That's why I ended up blending with the black plate. The magenta and yellow information is so corrupted, anything that toned it down, even it it pushes towards neutral, seemed to be an improvement.

Might there be a larger resolution version available? That would reduce the jpeg artifacts in relation to image information. Be one less layer of grunge to remove.

rutt · November 28, 2005

I was able to take a move or two further. I used selective color/red to add some cyan, remove some black, and generally cool it:

Now I was able to lighten it without making it worse than it alrady was:

All in all, I'd say this struggle is a good advertisement for a shoot-time solution to the problem, like the photographers really want. Of course this very shot can't be saved that way. No reshoot here! But film would have been the way to go here, it seems.

koopmans · December 1, 2005

edgework wrote:

Might there be a larger resolution version available? That would reduce the jpeg artifacts in relation to image information. Be one less layer of grunge to remove.

I made a public gallery available here with the full quality original

http://koopmans.smugmug.com/gallery/1008559/1/46677479

Also note that these are the original, unedited versions – so in some cases they are a bit dark and you have to brighten it before you see the redness. The other one that was posted had already been edited slightly.

Thanks,
Chris

edgework · December 1, 2005

Well, here's a better result than the last one I tried, but it wasn't pretty. Or quick. The large resolution image was easier to work with, however.

I have some fake sRGB profiles at reduced gamma settings, 1.8, 1.6, and 1.4 (easily made through custom RGB in the color settings). I assigned the 1.4 gamma profile which lightened things considerably, then converted to standard sRGB, to hardwire the lighter settings, and then repeated twice more with Colormatch RGB (1.8 gamma).

Then I duplicated the image and converted that to CMYK and copied the black channel back to a layer, in luminosity mode. At 31% with a mask for the face and hand, that helped smooth out some of the blotchiness.

For the rest, I treated it like the color problem it is, instead of trying to work it over with the healing brush. I sampled the skin tone that was decent, and a patch that was blotchy red. Then, in a curve adjustment layer, I simply set numbers for anchor points moving each curve from the dark value to the light. I filled the layer mask with black. Then, with a low opacity brush in pressure sensitive mode, I painted my curve into the red areas. It paints in an entire range, so the same curve produces good results even though there is variation in the blotchy areas. And there's no cloning, so skin texture is retained.

After that, I pretty much followed the lab formula for portraits. A quick stop back in RGB to add a little cyan to the Reds and Yellows (that portrait formula makes things too warm for my taste) and then on to CMYK for a luminosity move in the magenta channel, pulling the quartertones down slightly. Finally back to RGB where I used the healing brush to smooth over a few rough spots.

rutt · December 6, 2005

I sent this to Dan Margulis as a challenge and he wrote back:

Dan Margulis wrote:

John,

It's not a hard shot to correct if you just face the fact that the skin
imperfections are going to have to be taken out artificially. Just correct the
image as if the red blotches weren't there, like in the Chapter 16 recipe.

The red blotches are easily isolated in the A channel. I didn't try to correct
the whole image, but I did save a copy of the A channel and edited it so that
the blotches were highlighted. Then, load it as a selection, and in LAB apply
the whole image to itself in inverted Overlay mode at 55% opacity. You might
have to try three different settings for the L,A, and B overlays but my way
seemed to work with just the one. When I brought the file back to RGB all three
channels looked good.

Best,
Dan

I didn't follow Dan's instrctions to the letter in that I didn't try to do this after the ch 16 recipe. Instead I tried it immediately on the original image and it does do a lot.

Here is that A channel after Auto Levels and quite a steep curve:

I followed Dan's instructions and loaded this as a selection and applied the whole image to itself in inverted overlay mode, opacity 55%:

Result:

Not a bad starting place! What's going on here? The inverse overlay blend tends to push contrast and color in the opposite direction. Very light things become darker, very colorful things become more neutral. By doing this through a selection of the blotches, we limit this to just the blotches, making them lighter and less colorful.

After getting here, I pushed though the ch 16 portrait workflow, but made a couple of changes:

Followed Edgework's idea of using a false low gamma RGB profile to lighten up the whole thing.
Used blending options to lower the effect of the green luminosity blend on the blotches:
Actually flattened the L curve though the area of the face a bit to make it smoother.
Replaced conventional USM with surface blur. I wasn't working on the best version of this image and the skin has a lot of imperfections which may be jpeg artifacts or may be something else.
Used a greater HIRALOAM radius and amount than I might otherwise to try to regain some of the face shape lost with the L curve flattening.

Baldy · December 6, 2005

edgework wrote:

That's very impressive!

rutt · December 6, 2005

Yeah, it's awesome to have Edgework onboard, isn't it?

edgework · December 7, 2005

rutt wrote:

I didn't follow Dan's instrctions to the letter in that I didn't try to do this after the ch 16 recipe. Instead I tried it immediately on the original image and it does do a lot.

Which just shows that these are not steps carved in stone, but ideas to apply as needed.

The key to using this trick successfully is in the A channel mask to isolate imperfections. This face is a real problem since the discoloration spans a wide range of values. Isolating one area with a curve tends to plug up others.

You did a good job with the mask, though, the problem I've seen with this one is that any areas that are not blotched, yet appear as less than black in the mask, will be degraded by the inverse blend, both in color and contrast. Not sure how to get around this.

cdhames · May 6, 2007

I found this post looking up info on color correction and thought i'd give it a quick try after reading through some of the problems. i know the threads over a year old. just wanted to post my effort at working it. took about 30 minutes, mostly in the RGB workspace, although I did cross over into lab for color saturation, and into cmyk for skin-tone control:

jsncarrier · May 27, 2007

decided to mess with it
I decided to give it a go via an alternate method. I used Levels and Curves on the individual RGB channels, as well as some layers and masks. It took me about 10-15 mins.

Correcting those newborn shots

Comments