Fixing flat looking faces

rutt

Here is another trick from Dan Margulis. This one is very different from the CMYK contract enhancement + cast removal usuallly associated from Dan.

Here is a portrait that came out underexposed and missing depth in the face:



and here it is after photoshop postprocessing:



Here is what I did.

1. Start in RGB
2. Make a duplicate layer
3. Apply image, green channel to RGB composite of duplicate layer, resulting in a black and white image.
4. Change blending option of duplicate layer to Luminosity.
5. Flatten image
6. Move to LAB
7. Curves, steepen A+B channels, move light endpoint of the L curve a touch toward darkness.
8. Sharpen the L channel.
9. Back to RGB, Done!

This is a lot of steps, but not actually hard. No backtracking. it just works.

mrleo

Very nice technique. I will definately have to remember this one.

divamum

rutt wrote:

Here is what I did.

1. Start in RGB
2. Make a duplicate layer
3. Apply image, green channel to RGB composite of duplicate layer, resulting in a black and white image.
4. Change blending option of duplicate layer to Luminosity.
5. Flatten image
6. Move to LAB
7. Curves, steepen A+B channels, move light endpoint of the L curve a touch toward darkness.
8. Sharpen the L channel.
9. Back to RGB, Done!

This is a lot of steps, but not actually hard. No backtracking. it just works.

Rutt -

Could you outline step 3? You often refer to applying channels, and I have yet to figure out how to use that information (ie I have no clue how to do it!) If that should be in "regular" finishing school, by all means answer there (or point me to a previous thread which addresses it), but I would LOVE to know how to use this amazing processing information you so generously share with us all! Without some of these basic skills, I don't know how to apply it (kind of like wanting to make an escoffier sauce without knowing what "boil" or "sautee" mean!)
Thanks!

Shishu

divamum -

The idea behind steps 3 and 4 are to use a good B/W copy of the image as a layer in luminosity mode to enhance image contrast. Any good B/W copy would work for this layer, but for portraits, the green channel alone is often good enough (and obviously very quick ).

So the process is to duplicate your background layer, and use the Apply Image command (Image -> Apply Image) on this layer and apply the green channel of the background to the RGB composite of the duplicate layer in luminosity mode.

Hopefully this made some sense!

I first came across this technique in this thread: http://www.dgrin.com/showthread.php?t=22524

Shishu

I should add, this step alone seems to make such a difference in my photos that I simply cannot bring myself to use a Lightroom only workflow, even for snapshots.

rutt

I was going to add a reference to that thread. In fact the technique described here is just a nascent version of the one described there, which has served as my basic workflow for quite a few years. But in any case, using the green channel as a luminosity layer for flesh remains a key element. Same goes for B&W conversions. Consider just taking the green channel as a starting point for the B&W portrait. A more refined variant of this B&W technique is here.

divamum

Thanks both of you - that definitely helps!!

jfriend

I'm having a hard time appreciating the change in contrast because of the skin color. Is nobody else seeing a skin color that's way, way off here? Shouldn't the big things get fixed first?

rutt

Jim revived a really old thread here and the image is from before I understood how to get believable flesh colors. I stand by the idea of using the green channel for a luminosity blend (and maybe before I knew how to fucus my camera.)

Today, I'd adjust the color first. Look at the LAB portrait thread. Maybe I'll get around to redoing this one from scratch and posting the steps here. It needs more than a that green blend.

rutt

These days, I'd get a better result from this OE image of a very pale skinned person. First of all I might even get it correctly exposed and focused. I'd shoot in raw and be able to recover better. Fine.

But it's a treasured memory for a friend, so I thought I'd see how much I could do to save it with my current post technique without too much agony. This is a variant of Dan Margulis' picture postcard workflow, applied far beyond picture post cards.

So starting with this:

1. Adjust color to fix cool cast
   
   I used a curves layer with color blend to get the blue and green out of her hair and face. This makes a subtle change that would be hard to see by eye. You have to rely on measurements instead of your calibrated monitor. Doing this now will enable us to enhance color later without getting strange colors.
   
   
   
   Result:
   
   
   
   
2. Green luminosity blend
   
   Our old friend from the original post. I applied it once:
   
   
   
   That was such a good thing, I flattened the image and applied it again. (It surprised me the first time I found out this wasn't a fixed point.)
   
   
   
   
3. Adjust contrast with luminosity blend curves
   
   In a luminosity blended curves layer I did this:
   
   
   
   Result:
   
   
   
   
4. Save a duplicate for later, convert to LAB
   
   
5. Make a layer, apply image in multiply mode
   
   
   
   The result is horribly dark and colored, but we'll fix soon:
   
   
   
   
6. Use the image as a layer mask to apply less color and contrast to the darker parts of the image
   
   Add a layer mask to the layer with the multiplied image. Use apply image to apply the green channel of the saved image to the layer mask (experiment with various channels.) Result:
   
   
   
   
7. Blur the layer mask to recover detail in the image
   
   The reason this works is subtle. Don't worry, be happy. I used Gaussian blur at radius 40. Result:
   
   
   
   
8. Back to RGB, accepting the offer to flatten, and blend with the duplicate (pre LAB) version
   
   Scale back to taste. I liked 50%:
   
   
   
   
9. HIRALOAM sharpening
   
   These days I like the color action this introduces when done lightly in RGB:
   
   

Final:



Lots of easy steps. Takes my about 5 minutes. Much longer to capture all the intermediate states and photoshop screen shots.

Shishu

Rutt, I am a little curious about step 8. It seems to me that at the end of step 7, we have a colorful version of our image on the top layer with an intelligent mask, and below, our contrast enhanced image. Why not simply play with the opacity of the top layer, rather than all this fussing about with a duplicate copy of our image?

And thanks for all the discussion about the picture postcard workflow...I am really liking this method!

rutt

Shishu wrote:

Rutt, I am a little curious about step 8. It seems to me that at the end of step 7, we have a colorful version of our image on the top layer with an intelligent mask, and below, our contrast enhanced image. Why not simply play with the opacity of the top layer, rather than all this fussing about with a duplicate copy of our image?

And thanks for all the discussion about the picture postcard workflow...I am really liking this method!

That's a very good question. I do sometimes play with the opacity in LAB as you suggest. In converting to RGB, I flatten before layering. I also sometimes play with a curve of the layer mask. And sometimes I shapren in LAB. That final step is sort of a fail safe against a flatten move in LAB or conversion to RGB. A big point of this workflow is to learn it and not to think too much along the way to make it fast.

I'll make the flatten on conversion to RGB clear in the above post.

Stephan Photos

rutt wrote:

These days, I'd get a better result from this OE image of a very pale skinned person. First of all I might even get it correctly exposed and focused. I'd shoot in raw and be able to recover better. Fine.

But it's a treasured memory for a friend, so I thought I'd see how much I could do to save it with my current post technique without too much agony. This is a variant of Dan Margulis' picture postcard workflow, applied far beyond picture post cards.

So starting with this:

1. Adjust color to fix cool cast
   
   I used a curves layer with color blend to get the blue and green out of her hair and face. This makes a subtle change that would be hard to see by eye. You have to rely on measurements instead of your calibrated monitor. Doing this now will enable us to enhance color later without getting strange colors.
   
   
   
   Result:
   
   
   
   
2. Green luminosity blend
   
   Our old friend from the original post. I applied it once:
   
   
   
   That was such a good thing, I flattened the image and applied it again. (It surprised me the first time I found out this wasn't a fixed point.)
   
   
   
   
3. Adjust contrast with luminosity blend curves
   
   In a luminosity blended curves layer I did this:
   
   
   
   Result:
   
   
   
   
4. Save a duplicate for later, convert to LAB
   
   
5. Make a layer, apply image in multiply mode
   
   
   
   The result is horribly dark and colored, but we'll fix soon:
   
   
   
   
6. Use the image as a layer mask to apply less color and contrast to the darker parts of the image
   
   Add a layer mask to the layer with the multiplied image. Use apply image to apply the green channel of the saved image to the layer mask (experiment with various channels.) Result:
   
   
   
   
7. Blur the layer mask to recover detail in the image
   
   The reason this works is subtle. Don't worry, be happy. I used Gaussian blur at radius 40. Result:
   
   
   
   
8. Back to RGB, accepting the offer to flatten, and blend with the duplicate (pre LAB) version
   
   Scale back to taste. I liked 50%:
   
   
   
   
9. HIRALOAM sharpening
   
   These days I like the color action this introduces when done lightly in RGB:
   
   

Final:



Lots of easy steps. Takes my about 5 minutes. Much longer to capture all the intermediate states and photoshop screen shots.

I wonder how a warming filter or modifications to the color balance in RGB mode would work?

pathfinder

Just a note - using a B&W image to add add/control contrast in a color image, can be done in the Adobe's Raw Converter converter quite nicely with a RAW image as a Smart Object. Maybe I need to demonstrate with a new thread?

rutt

I'd really like to know how to get a green channel conversion in ACR.

pathfinder

I thought about this a little while this evening, John, and I do not have time to describe all the steps, so I gathered some screen shots that may help explain what I am talking about.

I chose this image of a neon Fat Tire sign since it clearly demonstrates the different RGB channels so nicely. Shot a couple years ago with a G10. Straight from a RAW file to Photoshop without any editing - using default RAW values entirely




You can quickly see the red, green and blue channels in Photoshop CS3 and earlier by typing ctrl-1, ctrl-2, and ctrl-3. ( I know you know this John , but some readers may not ) In CS4 you have to type ctrl-3, ctrl-4, ctrl-5. This is a good thing to quickly run through after importing your image into Photoshop to see where the contrast lies. Marc Muench teaches this routinely in his workshops.

John, your use of the green channel relies on the ease of access to the green channel. You can create an alternative B&W layer for blending by using Image>Adjustments>Black and White and creating any monochrome rendition you desire.

With Smart Objects you can do the same thing with the Hue and Saturation tools with the Convert to Greyscale box checked.

After opening your initial edited RAW file into Photoshop as a Smart Object by holding down the shift key while pressing the Open button the file opens in Photoshop at a Smart Object and has the little corner in the lower right of the icon turned up, indicating it is a Smart Object and thus is capable of being returned back to the RAW converter for further editing. We will copy the background layer as a new Smart Object by clicking on Layer>Smart Object>New Smart Object Via Copy. This creates a new Smart Object Layer that is not connected to the first background layer. Now we can click on the little turned up corner in the lower right of the icon on the upper layer and return the image back to the RAW converter.

Once we have the upper image layer back in the RAW converter, we can click on the HSL/Greyscale icon and also click on the Convert to Greyscale button. The image will turn to monochromatic immediately. Now we can alter the rendition of this greyscale image by sliding the sliders of the various colors until we achieve the luminosity rendition we desire.


Here is a screen shot of the red channel in Photoshop on the left, and a similar greyscale image from a Smart Object in the RAW converter on the right. I make no claim that these are identical, as I did not take the time to use the eyedroppers to verify that they are identical, but I think you will admit they are pretty close for a 30 second attempt to create a red channel greyscale image in ACR.




Here is the corresponding green channel and a similar creation in ACR



And the blue channel in PS, and my attempt to create a similar grey scale in ACR



Oce you have the appropriate greyscale rendition in ACR you simply click open image and the upper layer changes from the colored image to the newly rendered greyscale for further image ( not channel ) blending.

I find the technique invaluable, and use it frequently these days. It is a direct outgrowth of channel blending, but with the advantage that ACR editing does not lose or mangle pixels.

One confusing note about these images - they all have the red channel in their name because I did not rename each image as I saved it. But if yoiu look the the right of the name of the image in the Photoshop title you willsee that it is the red, or the green, or the blue channel.

Examine how the seat and the handlebars appear in each image and you will see that they are the appropriate grey channels.

I am genuinely interested in your opinion, pro and con, about this technique, rutt.

I am not aware of the precise math use to create the greyscale image in ACR, but I did test several areas with the eyedropper to see if they are similar in the grey scale jpg and the corresponding greyscale rendition in ACR.

rutt

You answered a different question than the one I had in mind. I keep running into people who want to do everything in LR these days. I'd like to be able to tell them how to do a green channel conversion. But perhaps this belongs in a different thread...

pathfinder

The speed and ease of LR are seductive, especially for shooters for whom time in front of a computer screen is money lost.

I use Lightroom, but only as a global editor. i have not developed much affection for the local adjustment brushes in LR yet. I still prefer to do my local editing in PS, and I find Smart Objects easier to use in conjunction with PS also.

Some images are fine with LR, but I find I can almost always make them a little better ( to my eyes anyway ) with a pass through local editing with PS as well. I wish I could do it all from LR too, but find, alas, I cannot.

I do not know a way to see the green channel in LR. Do you?

We can open a different thread, for the rest of this conversation,

rutt

pathfinder wrote:

We can open a different thread, for the rest of this conversation,

I asked the question in this thread, which I think is appropriate.