Generating Skin with Frequency Separation

[gene_spark]

Thats my first Thread here because i need a litte help.


I have found a littel trick that i used mostly for quick and dirty retouching and very bad skin.... and sometimes it works very fine and saves a lot of time...

when there is a good skin part i copied that part of the HF Layer to the Clippboard...
than New Layer on normal clipped with the HF layer...than I opend the Pattern Maker...use Clippboard as sample and generate a Pattern with Sample Detail around 10-20 ...

Than i put a black mask and paint ...

The Problem is that often the generated Skin is too soft for my taste...and looks blurred .
has anybody out there a little advice..
and puhhh I know my english :-))

[Nomad Jess]

Hi ,

I needed to recreate skin as I'm in the middle of restoring an old photo. I came across this tutorial and it worked really well. might be worth having a look at it : ) Good Luck

http://www.photoshopsupport.com/phot...-tutorial.html

[julianmarsalis]

Try clipping curve adjustment layer to your newly created layer. Generally can increase the sharpness by varying your curve make sure you anchor a point at 128x128 or you will change your 50% grey and lighten or darken the image overall.

[bakerser]

Are you by chance painting it in with a soft brush? This will invariably blur the sampled texture.

If not that, can you share a layered PSD sample (just a small crop even).

[bakerser]

Originally Posted by julianmarsalis

Try clipping curve adjustment layer to your newly created layer. Generally can increase the sharpness by varying your curve make sure you anchor a point at 128x128 or you will change your 50% grey and lighten or darken the image overall.

This works, but one should know that one needs points at both 127,127 and at 128,128 in order to prevent overall shifts (the unfortunate reality of Adobe not yet implementing floating-point curves). If superfund control of light / dark sharpening isn't needed, a Brightness / Contrast layer in Legacy mode can accomplish the enhancement without any global effects.

[Der_W]

Originally Posted by bakerser

This works, but one should know that one needs points at both 127,127 and at 128,128 in order to prevent overall shifts

Why so?
Working in 8Bit we have 2^8 = 256 tones per channel.
256/2 = 128, which is neutral gray.

You might think, that because 255 is white, it would be 255/2 = 127.5 (now your idea would make sense to me), but you're forgetting about the first value being 0 (which brings one more tone into play).

Or did I misunderstand you completely?

[bakerser]

Originally Posted by Der_W

Why so?
Working in 8Bit we have 2^8 = 256 tones per channel.
256/2 = 128, which is neutral gray.

You might think, that because 255 is white, it would be 255/2 = 127.5 (now your idea would make sense to me), but you're forgetting about the first value being 0 (which brings one more tone into play).

Or did I misunderstand you completely?

Actually if we had only 255 levels this would be better than having 256 (for this purpose only), as we need a value which PS can treat as being 'neutral'. We do, in fact, want a 127.5 so as not to do anything in those areas which have no 'HF' data. That's actually why PS's 16bpc mode is really 15bit + 1 level (so that there is a true neutral value).

[gene_spark]

Thank you all

@ Nomad Jess
One of the first Books i read about Photoshop was Katrin Eismanns first Edition of retouching & restauration and there she explained this burred Nois and embossing technik..and i used it sometimes before I knew the freq. sep Thing...and of course i tried it on the Hf Layer...but for me it looks too fake :-))

I tried The clip curve technik The tip with two Points is cool.thanks for that...and i tried to do the seperation with with a lower scale than 2 that gives allso a bit more sharping....

Has somebody may be an Idea to avoid the Pattern Maker?

[kav]

I wouldn't worry about 128 being off. It won't shift the values on overlay, linear light, softlight, etc. 50% grey used for the neutral fill when creating one of these layers as a blank layer is set at 128. Create as many of these layers as you like with the 128,128,128 fill on an 8 bit image, set them to a group and place color samplers on various points of the image. These samplers won't change when you click the group on and off, or if you duplicate the document and merge one while deleting the group from the other. Anyway it was just to prove that regardless of if it's a perfect middle grey it works in this kind of blending mode, and yeah the 0 does make it irritating.

[bakerser]

Originally Posted by kav

I wouldn't worry about 128 being off. It won't shift the values on overlay, linear light, softlight, etc. 50% grey used for the neutral fill when creating one of these layers as a blank layer is set at 128. Create as many of these layers as you like with the 128,128,128 fill on an 8 bit image, set them to a group and place color samplers on various points of the image. These samplers won't change when you click the group on and off, or if you duplicate the document and merge one while deleting the group from the other. Anyway it was just to prove that regardless of if it's a perfect middle grey it works in this kind of blending mode, and yeah the 0 does make it irritating.

True for 8bpc images (rounding 'protection'); not so for 16bpc images, where the image will suffer gradual creep one direction or the other. My caution was targeted towards the latter category, as I'd understood Der W's comments to be in regard to the Curves dialog's resolution -- if I was wrong there, please disregard with my apologies.

[kav]

Originally Posted by bakerser

True for 8bpc images (rounding 'protection'); not so for 16bpc images, where the image will suffer gradual creep one direction or the other. My caution was targeted towards the latter category, as I'd understood Der W's comments to be in regard to the Curves dialog's resolution -- if I was wrong there, please disregard with my apologies.

I haven't tested that, but on 16 bit images they use totally different numbers on this technique, and the 128 offset isn't put in. I haven't ever seen rounding creep personally. I am tempted to test this now. i'm not sure what 16 or 32 bit numbers photoshop uses for a middle grey approximation. Obviously 128 is still entered on the user end which is an 8 bit value. I'm getting 0-65535 as 16 bit numbers via simple math of 2^16 with the value for 0 taken out. If it's 15 bits that would mean 0-32767. I don't know exactly how these are distributed in 16 vs. 8. If it's an even distribution 32768/256 = 128, giving each 8 bit number 128 values that point to it from a 16 bit number if they're using this method of distribution which I'm not sure is the case. Since it's a midtone value it should be close enough for approximation. By this math showing a midtone creep via a blending mode that is considered to use middle grey as a neutral value, you'd need a lot of layers for any bit flipping in 16 bit numbers to move one of the three 8 bit numbers a full point. It's not done with a 128 offset at 16 anyway but via using a scaling factor of 2 and an inversion to approximate. I haven't been able to get any values to creep so far. I'll try doing this the separation thing + copy merged 10 times and see if I can produce a numerical shift on a few different points across the image. I could see this being more of a consideration something other than working on photographs for the purpose of enhance their look.

[bakerser]

To address a few things:
- PS absolutely does not offer 65k values. Whether this is a good thing or not is much debated - I happen to think that it is. In reality it offers 32769 values (0-32768), allowing a perfect neutral value at 16384. You can verify this by filling a 16bpc layer with 50% gray and reading it with the eyedropper.
- The problem with 8bit curves is exactly what you're getting at - with LL-family separations. Indeed, if I were just looking at a normal layer and adjusting it by 1/32k, it wouldn't be that big a deal for most all applications. But when I attempt to apply a curve (which interpolates values mind you - even our 'lock' points are not inviolable), that 1/2 point(+) distance from true 50% gray which the curve maps out becomes a full point move after the LL doubling is brought into play. This forces an undesirable move in the result where we shouldn't have to have one.
- There are of course lots of ways to deal with the above - it's just a minor limitation in truly fine work, and an inconvenience for the rest.

[bakerser]

A brief introduction to visual frequencies for those not yet accustomed: here.

Part Two

Part Three

[russpears]

What is the HF Layer, LL etc.

[bakerser]

Originally Posted by russpears

What is the HF Layer, LL etc.

HF = "High Frequency"
LF = "Low Frequency"
LL = "Linear Light"

Let me know if there are others which we should have been defining earlier .

[russpears]

What is a gradient "helping" layer used for? And how is it made?

[ray12]

If your base skin background is 127 or 128 you will not likely notice a very substantial effect. That kind of difference would only change the brightness level by a very little amount. It doesnt affect color itself since it is a neutral gray color...just the luminance. That mid color totally drops out with some blend modes...so it is not a big issue. Besides, the skin pore textures are painted in very slightly and are blended in...so again...very little effect.

Ray12