Isolate hair

[lacogada]

Have an image with different shades of gray background.
Trying to isolate girl and hair.

Duplicated background.
Placed red layer between the 2 image layers so transparency would show.

Color to transparency with a tolerance of 3.
Still removes much of skin and hair.

Looking for tips / suggestions on how to process.

https://spaces.hightail.com/space/7zDU4 ... arency.jpg

[lacogada]

If you would like the full image.:https://spaces.hightail.com/space/7zDU4 ... 1/girl.jpg

[Hoogo]

Quite evil...
The background has a gradient, and the hair has it all: Highlights, dark strains, and semi-transparency.

If you shoot the photos yourself: Take a photo of the empty background with the same settings.The difference between those pictures can be used to create a mask rather directly and easily, but some hair will be missing.

I think I would create a few different layers with different tools:
- Autolasso with a little softening for the shoulders. Quick selection might work very weill, too.
- A mask created with levels for the parts of the hair that are surely not transparent.
- Surely a layer with semi transparent hair, separate layer to multiply into the result
- Another layer with the highlights in the hair on top ( or included into the first layers)

Not sure if levels will do for the last 2 layers, maybe. Or maybe creating some fake empty background to help selecting them.
Will need some experiments...

[lacogada]

Yes it's a tuff one.
I did not take the photograph.

Thanks for your thoughts and suggestions.

[Vivi_Ram]

The way I deal with it, is by going to see which channel has the maximum contrast between background and subject. Then, channel to layer it. Then, add a levels adjustment to this new channel layer, to increase the contrast until the darker parts are nearly completely dark. And brighter spots are nearly completely white. Then, I take brush and whatever wasn't covered by levels, I adjust manually. Painting black over midtones that were missed by levels. And white over background missed by levels. Now, I apply maximum adjustment to see if I can get some of the softer edges back into the hair. Then, applying gaussian blur adjustment to further soften the hard edges caused by maximum. You should have a fairly manageable mask now. You click on the layer and turn into Mask for Previous. Let me know if this helps.

[lacogada]

Thanks try to follow but not sure of some parts

[Hoogo]

The way I deal with it, is by going to see which channel has the maximum contrast between background and subject. Then, channel to layer it.

If you want to extend that idea of the channels:
- Use colorpicker and get 1 "typical" RGB value of background and object. Or the RGB values of a more difficult transition.
- Subtract them on paper.
- Go to the cannel mixer and enter the resulting RGB values to create a grey output, May need some fiddling, doubling/halving the numbers, some play with the constant...

That way you have created a part of your own, special and perfect color space, one channel that divides the 2 selected colors the best.

That's the "quite evil" part: The created mask is nice only for some parts of the picture. Other parts of the picture need different masks. And still, some parts of the hair will still be invisible.

"...how to paint on it..."
- Well, painting on the adjustment layer won't give the expected result.
- Painting black or white on the red layer will work, and painitng black makes sense at some areas. But not everywhere, and surely that would be a bad idea on all the areas with bright skin.
- You will have to combine different masks. A good start is this:

Mask of layers.png

- Create a group,
- move the mask elements to that group,
- add other elements of the mask to the group. Here I just added a simple layer for painting. For combining the mask elements, some layers can be "screnn", "darken" or whatever helps for combining.
- Set the group to "encapsulated" with the layer attributes dialog, that's the Asterisk * in the group name. In your current setting it doesn't make much of a difference, if at all, but it won't hurt.

[Vivi_Ram]

You've missed a step in that description. We still have to paint black over the face completely on the channel layer and once complete, you'll have a completely black. subject and completely white background. Now, you can invert the layer. Now you have a layer that you can turn to a mask. Incidentally, it's a completely non destructive mask. You can even add a second mask on top. And it will "add" the two masks in result.

Let me share an example later today.

@Hoogo's suggestion should work too. More than one way to skin the cat.

[Vivi_Ram]

@Hoogo: Keen to learn your suggested method. See my question below.

If you want to extend that idea of the channels:
- Use colorpicker and get 1 "typical" RGB value of background and object. Or the RGB values of a more difficult transition.
- Subtract them on paper. - Could you please explain this step? What do you mean by subract them on paper?
- Go to the channel mixer and enter the resulting RGB values to create a grey output, May need some fiddling, doubling/halving the numbers, some play with the constant...

[Vivi_Ram]

@Martin & @Gerhard or anyone else: https://www.youtube.com/watch?v=SVqJOny0FyQ - is this method possible in PL. If yes, where would I find it. It is extremely useful where you have to make very detailed selections that are impossible in any other way.

[Herbert123]

@Martin & @Gerhard or anyone else: https://www.youtube.com/watch?v=SVqJOny0FyQ - is this method possible in PL. If yes, where would I find it. It is extremely useful where you have to make very detailed selections that are impossible in any other way.

This can be achieved non-destructively in PhotoLine. But requires a very different workflow - or at least the same core approach, but working with layer duplicates, layer masks, and the Channels in the layer properties.

If I have the time, I will whip up an example.

[Vivi_Ram]

This can be achieved non-destructively in PhotoLine. But requires a very different workflow - or at least the same core approach, but working with layer duplicates, layer masks, and the Channels in the layer properties.

If I have the time, I will whip up an example.

I'd be very grateful if you could. Thanks!

[lacogada]

Thanks for all the replies and info guys.

I really struggle with PL ... guess I’m just to use to the way my old PS CS6 handles layers and mask.

[Herbert123]

This can be achieved non-destructively in PhotoLine. But requires a very different workflow - or at least the same core approach, but working with layer duplicates, layer masks, and the Channels in the layer properties.

If I have the time, I will whip up an example.

I'd be very grateful if you could. Thanks!

I had some time on my hands between work just now, and I 'cloned' that method in PhotoLine. WIll share it later tonight with you.

[Vivi_Ram]

@Herbert123: looking forward to seeing it. Thanks!