Frequency Separation -- various techniques?

[bkh]

One quick thing I've noticed about the 3-way VC FS action is that it slightly darkens more of the image pixels than does the PL FS action. In other words, the PL action slightly darkens only the most dark pixels, thus slightly increasing the contrast in the dark areas (such as the iris), which is a Good Thing. This Virtual Copy action slightly darkens a wider range of pixels, thus slightly decreasing the contrast which is Not Good.

As I wrote, if 8 bits gives errors, go to 16 bit. Unlike the PL version, deviations from the original are solely due to rounding errors. In my experiemnts, the result of putting the three layers together in 16 bits is indistinguishable from the original.

Currently, when you use the 8 bit version on a grey layer, the MF and HF layers are at value 127 each instead of 128 (the neutral point of Linear Light). That's probably the cause of the general darkening you observe. But the PL version has the same problem. I'll see if that can be fixed easily.

Cheers

Burkhard.

[cathodeRay]

Not sure why there is such an apparent reluctance to go to higher bit levels. 8 bits has inherent problems like no proper mid-point, 16 bits are better but there are still clipping problems but 32 bits pretty much takes care of everything (including allowing the use of simple subtraction rather than 'add an inversion'). File sizes are potentially much larger but modern storage is affordable, and since many of the layers are virtual copies the file size does not increase too much anyway.

Nor am I inclined to think that editable text rather than binary actions would cause a significant performance hit. When things are slow in PL, it is because an awful lot of behind the scenes calculations are going on (a complex filter, for example). It's the calculations that take the time, not the reading of a text script. I do still find the current actions system clunky - a bit like doing surgery by hitting the outside of a body with a hammer. Text scripts/actions would allow proper surgery - quick incision, do what's needed, close up and out again.

cathodeRay

[bkh]

That's probably the cause of the general darkening you observe. But the PL version has the same problem. I'll see if that can be fixed easily.

Here's the improved version.

Cheers

Burkhard.

[photoken]

Here's the improved version.

OK, thanks! I'll use this one for testing.

Just so I understand how to use it, I've got some questions:

1. To modify the High Frequencies and Middle Frequencies groups, I need to show each one in "Normal" blending mode while changing the blur radius in the Low Frequencies group. Correct?
2. In order to do any retouching, I must first reduce each of the frequencies groups to a single image layer using "Merge Layers" on each in turn, working from High Frequencies to Low Frequencies. Correct?

[bkh]

Just so I understand how to use it, I've got some questions:
To modify the High Frequencies and Middle Frequencies groups, I need to show each one in "Normal" blending mode while changing the blur radius in the Low Frequencies group. Correct?]

Adjust the blur radius in the low frequencies to separate low from middle frequencies, and the one in middle frequencies to separate middle from high frequencies.

It's probably easiest to use "show active layer only" in the context menu of the "eye" (or just alt-click the eye) to show the low/middle/high frequency layers only. In this way, you don't have to change the blending mode. Alt-click again to show all layers.

In order to do any retouching, I must first reduce each of the frequencies groups to a single image layer using "Merge Layers" on each in turn, working from High Frequencies to Low Frequencies. Correct?

Yes.

Cheers

Burkhard.

[cathodeRay]

It does seem to me that routinely introducing a third (of for that matter even more) separation layers is not so much the Third Way and the convoluted way. If the manual recipe you have to follow after setting up the stack via an action runs to pages and involves much merging (a destructive process) of layers and the risks of blowing out virtual layers while they are still needed because you went at the stack from the wrong end, then perhaps, unless three (or more) layers are absolutely necessary, go for simplicity (two separations), use 32 bits (so the maths works), keep the stack intact (don't merge) and use adjustment layers for retouching wherever possible. The dynamic visual feedback on the HF layer as you adjust the LF blur as you set up the separation (which IMO is a key part of the 'breakthrough' attached to these methods) is readily available. If you do later want to undo something, no problem. On the other hand, with a three way separation with multiple merged layers and destructive retouching, there's a considerable risk of one wrong move later on in the process, and you have to start all over again, right at the beginning.

[photoken]

with a three way separation with multiple merged layers and destructive retouching, there's a considerable risk of one wrong move later on in the process, and you have to start all over again, right at the beginning.

I don't understand that objection. Removing a defect/blemish is an all-or-nothing affair -- the blemish is either gone or not. You see the result right away, and if a retouching step produces a bizarre result you're instantly aware of it and "Undo" (Ctrl+Z) immediately sets things right....

[photoken]

It does seem to me that routinely introducing a third (of for that matter even more) separation layers is not so much the Third Way and the convoluted way.

Actually, the PL Frequency Separation action is primarily a two-layer method. The High Frequency layer is limited to very fine details -- in portraits that means things like the eyelashes, hair, the line separating the lips, etc. These are areas of a portrait that are rarely involved in removing blemishes, so the High Frequency layer will almost always not be used.

You can see this by applying the action to a portrait, changing the blend mode of the HF layer to "Normal" and applying a Threshold adjustment to the layer with a setting of 49%. The affected details will be revealed as black areas against a white background.

[cathodeRay]

I just like a non-destructive workflow. Maybe I'm just clumsy! Sure I can Undo the most recent retouch, if I don't like it, but because of the linear sequential nature of Undo, once I've added some other retouch that I do like, but want to go back to Undo something earlier, things as we know get more complicated. With adjustment layers, I just re-edit the relevant area. In the example below, I ended up doing more HF than LF retouching.

Taking the photo of the girl we started with, lets say (for demonstration purposes) I've examined the original and want to remove some of the lines under the right eye, remove some of the yellow tinge around the mouth, lower the high spots on the nose and cheek and apply a light general de-pore, if that's the word I'm looking for. Here's the actual before (above) and after (below), with the stack visible. The LF retouch was done using a light colour wash using a colour pipetted off the image, the HF retouch most with the copy brush and/or the paint brush set to neutral grey, all settings set to create a gentle effect. The visual check layer at the top is the easy one-click before and after check. All very quick, easy and low stress:

Pre-retouch.jpg

Post-retouch.jpg

The only thing I haven't worked out how to do non-destructively (and even destructively isn't easy), as I mentioned before, is how to add contrast, as in, say, adding some 'digital mascara' to say the right eye lower eyelashes. I haven't been able to beef up the eyelashes non-destructively without affecting the underlying skin (most of what I tried used the Copy Brush with various settings, then I tried direct editing of the HF layer with the Filter brush, ouch!).

cathodeRay

PS having just done a preview of the post, I notice the after image above is a little flatter/more desaturated, even in areas I didn't retouch. This isn't the case when I look at the images in PL - some sort of upload/forum image handling artefact?

[Hoogo]

The only thing I haven't worked out how to do non-destructively (and even destructively isn't easy), as I mentioned before, is how to add contrast, as in, say, adding some 'digital mascara' to say the right eye lower eyelashes.

What about a sharpening adjustment layer in layer mode "darken" between HF and HF adjust?

[cathodeRay]

Hoogo - I tried that, slightly differently, by using a Filter Brush set to sharpen. The problem I had was the sharpen effect behaved as expected, and increased the contrast 'both ways' and so introduced a white matte surrounding effect as much as it darkened the eyelashes. Having researched 'eyelash enhancement' etc it seems most established techniques rely either on selecting/copying and distorting the existing eyelashes and/or using an eyelash brush and/or vector drawing, in other words all are really creating new eyelashes in some shape or form, not enhancing the existing ones. It's also extremely tricky selecting eyelashes - perhaps one of the masking plugins might do it (I have Corel's KnockOut but it struggled). Alternatively, the fact the eyelashes are HF may be a red herring - maybe an inverted luminosity mask, perhaps based on the red channel, might be the way to go. But then that might be a red herring too...

cathodeRay

[Hoogo]

Like this?

[photoken]

lets say (for demonstration purposes) I've examined the original and want to remove some of the lines under the right eye, remove some of the yellow tinge around the mouth, lower the high spots on the nose and cheek and apply a light general de-pore, if that's the word I'm looking for.

Now I understand how you're working, and, yes, trying to do all that on FreqSep layers makes it desirable to have non-destructive editing, a smaller number of layers, etc.

I don't work that way -- I use FreqSep for removing blemishes (the "lines under the right eye" in your example) only. Once I've done that, the FreqSep layers are no longer needed so I'll flatten the image and continue the other retouching non-destructively....

[cathodeRay]

Hoogo - nice one, it works. For others trying to do the same thing (well, you never know), I had to add another normal (not adjustment) blank transparent layer and then convert (modify) it to be an adjustment layer - otherwise it kept on being formed as a child layer. It also arrived with an 100% full (black) mask but that was useful as it meant is was all set up to use white paint to reveal the areas where I wanted the 'mascara'. As ever, gently does it with this sort of retouching.

Ken - glad you now understand where I am coming from, even if you your a different approach. I've seen some eyelash stacks of over 20 layers though and I am not in a hurry to emulate that! More is less...

cathodeRay

[Hoogo]

Hoogo - nice one, it works. For others trying to do the same thing (well, you never know), I had to add another normal (not adjustment) blank transparent layer and then convert (modify) it to be an adjustment layer - otherwise it kept on being formed as a child layer.

If you use the dialog "adjustment layer", then it is an option whether a child or a normal layer is created.
I prefer to use the normal "filter > sharpen > unsharp masking" and then press the button to the left of OK to create a layer. Then double-click in the layer list to get the real preview with the usual filter sliders.

Another tip for this sharpening layer:
-There's that button with the red blotch beneath the preview. Hold shift while pressing it to get a preview that shows differences created by the filter.
-Start with intensity 500%, threshold zero.
-increase size until everything you want to sharpen appears.
-Increase threshold until all noise disappears again.
-press the red button with shift again.
-decrease intensity until you like it.

EDIT: I wonder if it really makes a difference if HP is sharpened or if the sharpening is applied to the whole picture.