The driving force behind this method is preservation of previous elements in multiple layers to allow easy re-editing in case an error is overlooked, or in case a client wants something small changed (no matter how small, it's likely to be easier to make a change in a layered file).

This method bloats the final file size (base of ~375MB for a 16-bit 13Mpxl image for me) but I feel this is an acceptable trade-off for the control it offers. I was recently able to go back and quickly unsharpen several files, edit, and then re-sharpen to adjust an issue that was brought up after file delivery, all in about 20 minutes.

Basically what happens is the image is split into 2 layers; a detail (high-frequency) layer and a blurred (low-frequency) layer. You can then edit the detail in an image without affecting the underlying structure.


Actions

http://maskphoto.com/files/frequency/Ma … y_trio.zip
Updated version! No more USM at the beginning, no needless noise, removed noise on top of healing layer where it belongs, and sharpen action behaves correctly with multi-layer composited source images.

2012-June-03: Newly Updated: 3 levels of sharpening as the final step, general clean-up, more accurate re-layering, more friendly healing.

There are 3 actions:

"Workspace" groups everything and sets up a liquify, and stops there (so you can examine the image and make decisions on further liquifaction).

"Frequency Split" (2 versions; 8-bit and 16-bit) splits the image into blur, detail, and noise (it's after this point that the true retouching begins). ALERT! you must either turn on ("check") "noise ninja" or "reduce noise" in this action, or add your own noise removal, or NO NOISE will be removed

"Final Effects" performs the creative sharpening mask, and optional local contrast boost and "grit" layers and should be done after all edits have been performed (though it's reversible: edits can still be done later).

Explanation of Layers

Running the "Workspace" action will give you this:


For landscape, fine art, or anything that doesn't require meticulous healing, this will be sufficient for your work.

From top to bottom, the layers:

+ Original: hidden, locked group containing all of the original image components (background and any compositing elements)
    bg: the original background layer (or layers, if you're working from a composite. Note that if you don't have a "background" layer in your original composite, the action will grumble at you)

+ Adjustments: contains all adjustment layers and dodge/burn/lightpainting "tone" layer.
    Darken: Rarely used; if burning on the "tone" layer is making the area too saturated, try painting the area's color on this layer; it's set to "multiply".
    Lighten: Rarely used; if dodging on the "tone" layer is making the area too desaturated, try painting the area's color on this layer; it's set to "screen".
    Tone: I use dodge and burn on this layer to dodge and burn ("carve") the image in a reversible manner. Also, painting in "color" mode can tone the image.

+ Retouching: this group holds large-scale retouching such as airbrushing solid colors onto a background. If you're not doing frequency separation, all your healing will be done here. When I do frequency separation, I almost never use this group for anything but "smoojing" in hairs via Natalia_Taffarel's method.
    Edits: a blank layer that's ready for retouching.

Base: This is a copy of the original image; do large-scale edits to this, such as liquefaction, lens correction, etc. Try to keep your localized edits on their own layers, to make re-editing a little easier if you ever have to revisit the file.

Running one of the "Frequency Separation" actions will give you this:


If you use an 8-bit file, use the 8-bit version. Likewise for the 16-bit version, as each uses a different method of frequency separation to rebuild the image.
NOTE: Each action will convert the image to the correct depth, so if you run the 8-bit on a 16-bit image, it will be irrevocably reduced to 8-bit. If you run the 16-bit version on an 8-bit image, the file will bloat, and you won't get any extra information out of it, so watch what you're doing.

Noise: A masked layer of the removed noise. Sometimes it's desirable to add some noise back in, such as in textured clothing areas. update:This layer can also be used to bring a little noise into an area that's suffering from banding, such as a blue sky that's been blurred out. Simply paint white or gray on the mask to bring some noise back in.

+ Detail: Set to Linear Light, this group holds all of the detail layers
    Capture: A masked curves layer to provide a slight (reversible!) amount of capture sharpen. Adjustable to taste.
    exaggerate contrast: A throwaway layer that I use during healing to make it easier to spot problem areas in skin texture, because I heal with the Base group turned off. Hide this when healing is completed.
    Healing: This is the layer I clone the High Frequency layer to, to allow reversible blemish retouching. You can either a) ignore this layer and clone right on the high frequency layer, b) set your clone source on the HF layer every time you need to adjust your source, or c) hide the base layer, set your clone stamp to "current and below", and use the "exaggerate texture" layer to help you see what needs to be fixed. "a" and "b" will let you see the healing in real time (though "b" requires lots of clicking), while "c" is fast and reversible, but you only get to see the abstract detail later while you work. update:This layer can also be used to mask in blurring, if desired; just airbrush 50% gray onto it and it will bring out the blurred "base" layer.
    High Frequency: one of the main parts of the image; I never edit this in order to preserve the original image data.

+ Base: This group holds the blurry base colors of the image.
    Smoothen: This layer is for smoothing out blotchy areas with an airbrush and otherwise altering the base color of an area.
    Bandstop: This is a copy of Low Frequency, but with an exaggerated blur and edge definition. It can quickly solve some blotchy areas by painting white into the mask.
    Low Frequency: this is the other main component of the image, and is never edited to preserve the original data. This is the blurred area from the GB step in the "Frequency Split" action. In that action, choose the lowest GB that still eliminates the blemishes you want to retouch.

Running the "Final Effects" action will give you this:


This action runs each of 3 separate actions, all of which provide "sharpening" at a different radius. "Sharpen" is small, and enhances edge definition. "Local Contrast" is large, and enhances local tonal differences, and "Grit" is medium, and provides a look similar to the Dave Hill effect. Grit is hidden by default.

You can also just run one or two of the 3 final actions separately.

+ Sharpen: this group is only created after all edits are done. It is blended with Linear Light and has an edge mask to allow "painting sharpness" where it's needed.
    Amplify: this curves layer amplifies the amount of sharpening. this can be adjusted to taste, or to sharpen highlights or shadows more or less.
    Diff sharpen: the sharpen layer. this is never edited.

Local Contrast: Masked aggressively to just provide a slight bump; edit the layer mask to taste.

Grit: Hidden and masked; unhide and edit the layer mask to taste.

Re-editing

Hide Final Effects, and perform your edits. Then (if you've modified the sharpen masks) change the names of the layers in the Final Effects group, and run the action again. Replace the masks in the new group with the masks in the old group and delete the old group OR duplicate the image to retain the old masks and delete the sharpen group, then replace the new masks with the ones from the duplicated image.

This allows you to retain the creative sharpening (masking) decisions you've made while being able to edit what happened before the sharpening was applied.


Notes

* It's annoying to clone from the High Frequency layer to the Healing layer, because you have to click multiple times to set your clone source, so I hide the base group and use the "exaggerate texture" layer and clone stamp with "current and below" to heal texture.
* if you split the frequency and later have to liquify or perform some other large distortion, you'll have to do it at least twice: once for the detail layer, once for the blurred layer, and maybe once for each masked adjustment layer you've created. You can save a liquify mesh and load it for each change you have to perform, but it's better to get it all done prior to the split.
* Removing large elements is occasionally not as easy as small elements; removing a blemish is easy. Removing a blob of sensor dust is not quite so easy.
* Removing some large elements is MUCH easier. I can iron out wrinkles in clothing and remove things like power lines and framed photos on walls very easily now (though it requires an edit to each layer, it's still better than before)

Group All Layers
The first step is to combine all of your current image elements into a group so they all get caught by the frequency split. At this point, all *I* have is the original or the image composite elements (I do compositing before just about anything).


Create Base Layer
This layer is what will get liquefied and ultimately split into detail and blur layers.


Create Edit Groups
The Retouching group is where i put large graphical edits such as clothing tag removal or airbrush-fixing a background to a solid color. The Adjustment group is where my dodge/burn layer ("Tone") goes, along with all adjustment layers.


Liquify/Lens Correction
Do any final gross edits such as figure enhancement or perspective correction.
After this point, any big edits have to be done twice (or more); once to the detail later and once to the color layer.

Action 2:

Noise Reduction
This step separates the noise from an image and places it (the noise) on a separate layer, masked to not display. I've found that noise reduction can reduce detail in busy areas such as grass or fabric patterns, so i feel it's important to be able to mask it back in where desired.


Frequency Split
Using the same method as the noise reduction split, this moves the detail from the image to another layer.


Create Detail Group
This sets up the detail layer to be healed in a robust manner, and performs a slight (reversible! and applied to all levels of detail) capture sharpen that can be adjusted to taste.


Create Base Group
You are now left with a multi-layered file that looks very close to what you started with. You may unhide/hide the "original" folder for a before/after view at any time during your processing.

Additional Processing Steps
It's fairly important to put things in the right groups, as some of the groups don't have normal blending modes and may not behave as expected.

Action 3:

Creative Sharpen
After ALL of your edits are done, creative sharpening allows you to add definition (clarity?) to areas that need it. This process creates a hyper-sharpened later and then masks to only the areas of high detail, BUT I almost always mask the skin out more and the eyes and jewelry in more. You'll still need to output-sharpen for whatever end-use you have in mind.

Edit: some values may change, based on discussion this thread seems to be engendering. Updates to the action will follow any conclusive feedback from the community. Thanks for reading, guys!

remerrill wrote:
Thanks for putting this up... this is wonderful!
I've bookmarked for later study...

+1
Me too.

SRB Photo wrote:
You weren't kidding about doing a write up!

And while I hate to drag this into the technical, I do have a few questions about the proposed workflow:

- Why a USM in the Workspace action?
- Why "Hard Light" for the noise reintroduction? (vs. LL)
- Why add noise to the LF data?
- Why a Levels adjustment of the HF data?
- Why midpoint your sharpening curve @ 128,129? (vs. 128,128)
- Why is the HF data curved by default (mask filled @ 25% gray x 80% blend)?
- How did you arrive at 3/5 * MP?

I'm curious .

I was wondering this as well. Please do explain, as you've piqued my interest this morning.

SRB Photo wrote:
- Why a USM in the Workspace action?

ack. forgot to include that. That was a holdover from some of my old processing. It does a slight bump of local contrast and is generally pleasing. I believe i picked it up from:
http://www.luminous-landscape.com/tutor … ment.shtml
It's also not technically reversible, but I found myself not worrying so much about that for some reason. I wonder if there's a way to handle that with a layer via Apply Image.
To be honest, i forgot it was even in there; it was late when i wrote some of that.

SRB Photo wrote:
- Why "Hard Light" for the noise reintroduction? (vs. LL)

I spent about 3 hours comparing noise and before/after layer application and "hard Light" was the only thing that seemed to return the image to normal after the noise removal.
It's possible i was working with flawed samples, but I tried 3 or 4 noisy images and linear light didn't seem to do it.
Now that you mention it, it *may* have been the fact that the noise layer was inside a linear light layer or that it was applying it directly to a split HF layer, instead of to the source image, but that might be my imagination (not knowing the specifics of the math behind the blending). It might also be the tweaks I did to the image copy layer with USM; i might have been correcting for something.

SRB Photo wrote:
- Why add noise to the LF data?

I found some disturbing banding show up in areas of similar color, and the noise reduced the effects of the banding without greatly affecting the quality of the image, visually. I suppose it could come out...?

SRB Photo wrote:
- Why a Levels adjustment of the HF data?

ummmmm.
Now we're into the tweaks that my math brain was turned off for. I *think* i just visually compared several before/afters and found that it was more convincingly "the same" if I bumped the midpoint slightly...?
This is probably related to my USM madness near the beginning.

SRB Photo wrote:
- Why midpoint your sharpening curve @ 128,129? (vs. 128,128)

see above?
Now that there's all this scrutiny, I'll have to revisit my actions and try to figure out why I did what I did.
(i made these months ago and quit photography in between, so I'm just a tad hazy. i guess i should have figured it all out again before going public.

SRB Photo wrote:
- Why is the HF data curved by default (mask filled @ 25% gray x 80% blend)?

Do you mean the Capture adjustment layer?
I found myself continually performing a capture sharpen. i know there is a lot of debate as to whether that's good practice; i side with the pundits who suggest starting with the optimally sharp version of your image prior to processing. I was pretty happy to find a way to do this in a manner that lets me turn it off and/or adjust it all the way up to final output. I can adjust my capture sharpen right before I perform creative sharpening if desired (unless a lot of compositing has been done in the Retouching group) - not that i've ever done this, but the option exists.

If you're wondering where I got my settings, this could be my amateur action scripting talking, but I couldn't reliably find a way to fill with a specific color, so I used a combination of brightness and blend to fill the mask (i wanted a mask so its opacity could be adjusted later).

SRB Photo wrote:
- How did you arrive at 3/5 * MP?

I did lots of playing around to find a good value for my images, but during the write-up, I realized that the value would have to vary based on the size of the file. and then i did my math wrong. ;-\ I actually think I used 1/1000 of the longest image size as an arbitrary starting point that produced pleasing results.

Koray wrote:
Funky how organized this looks when compared to my mess of layers and saved versions

oh yes, i'm very familar with messes of layers. I ended up running into issues where i couldn't find things, or would forget a step. Having it all set up at the beginning helps to alleviate that.

Koray wrote:
The problem with single separation (that I see alot here) is that once you smooth the LF the whole image looks blurred and makes me cringe for real. Nothing but a fancy way of old blurred look.

are you sure you're looking at it right? yes, the LF layer is blurred, but the HF layer brings back ALL of the detail that was lost.

I've tried multiple separations but just ended up doing the same retouch twice; i saw no benefit from multiple separations.

Mask Photo wrote:
are you sure you're looking at it right? yes, the LF layer is blurred, but the HF layer brings back ALL of the detail that was lost.

I remember reading surface blurring LF and call it smooth?
found it you call it bandstop.

when I mean multiple I mean separate once, do things, merge visible to a new layer, separate again, proceed.

I usually go for one large radius and then one small.

Koray wrote:
I remember reading surface blurring LF and call it smooth?
found it you call it bandstop.

ah yes. the bandstop layer is completely masked out (and only applies to the already blurred base layer in any case). I mask it back in to deal with areas of blotchiness or uneven color in a model's skin. you can achieve a very similar effect by just airbrushing on the Smoothen layer, but the bandstop is usually very close to the right color (and does a bit of edge preservation, to prevent color bleed from the high level of blur).

it's definitely not a standard surface blur; all the detail is still preserved in the HF layer in the group above.

This workflow might be strange to contemplate if you do multiple steps in sequence, and flatten in between, but I find it useful because I can change something from any step of the process, in any of my images from the last... ~year or so? I used to retain stacks of merged snapshots (you can actually just save history states instead of saving whole merged layers), but I found I was unable to really change something from the beginning without going through and re-doing everything that came after that one change.

edit: i think i see what you're saying now; you'd do one separation and take care of the big things, and then another separation to do the small stuff.
yeah, that would probably produce a better result. I'm a little too impatient, and i guess i feel i can solve the big things and small things at once with one HF and one bandstop.

Koray wrote:
I'm not questioning you but the others.

Alot of people just blur the LF and think they can get away with since the detail ls still there.

Looks awful especially when resized for web.

oh, for sure. if you were to unmask the bandstop completely, it would look hideous. all of the subtle tonality of the face would be gone and edge areas would have the wrong lightness

Mask Photo wrote:
edit: i think i see what you're saying now; you'd do one separation and take care of the big things, and then another separation to do the small stuff.
yeah, that would probably produce a better result. I'm a little too impatient, and i guess i feel i can solve the big things and small things at once with one HF and one bandstop.

I have three steps; cleaning, retouching, contouring.

After I'm done with cleaning I already have an idea about what the other problems are.

Then I separate targeting for my approach on how to fix those problems kinda guessing what goes up and what stays down.
Thats why I usually need more than one separation most of the time.

Mask Photo wrote:
Thanks for the questions. I'll try to hit them quickly now and i can discuss more later (if there's anything that needs discussing)

...

I spent about 3 hours comparing noise and before/after layer application and "hard Light" was the only thing that seemed to return the image to normal after the noise removal.
It's possible i was working with flawed samples, but I tried 3 or 4 noisy images and linear light didn't seem to do it.
Now that you mention it, it *may* have been the fact that the noise layer was inside a linear light layer or that it was applying it directly to a split HF layer, instead of to the source image, but that might be my imagination (not knowing the specifics of the math behind the blending). It might also be the tweaks I did to the image copy layer with USM; i might have been correcting for something.

Very likely your final explanation is correct.  You can either clip the noise layer to the HF data in LL, or change the HF data to LL and the group to Pass-Through.

Mask Photo wrote:
I found some disturbing banding show up in areas of similar color, and the noise reduced the effects of the banding without greatly affecting the quality of the image, visually. I suppose it could come out...?

10:1 you're just seeing the effects of PS's previewing engine when you talk about banding.  Replicate the situation, and then stamp visible above everything to see if the banding is really there.

Mask Photo wrote:
ummmmm.
Now we're into the tweaks that my math brain was turned off for. I *think* i just visually compared several before/afters and found that it was more convincingly "the same" if I bumped the midpoint slightly...?
This is probably related to my USM madness near the beginning.

If you want to be hyper-accurate with your HF curve clamping, a 50% opacity 1pt curves shift before the primary curve is the way to go about it.  It's still not as perfect as new curves would be, but it's as close as we can get while still using curves to boost the contrast.

Mask Photo wrote:
Do you mean the Capture adjustment layer?
I found myself continually performing a capture sharpen. i know there is a lot of debate as to whether that's good practice; i side with the pundits who suggest starting with the optimally sharp version of your image prior to processing. I was pretty happy to find a way to do this in a manner that lets me turn it off and/or adjust it all the way up to final output. I can adjust my capture sharpen right before I perform creative sharpening if desired (unless a lot of compositing has been done in the Retouching group) - not that i've ever done this, but the option exists.

If you're wondering where I got my settings, this could be my amateur action scripting talking, but I couldn't reliably find a way to fill with a specific color, so I used a combination of brightness and blend to fill the mask (i wanted a mask so its opacity could be adjusted later).

I don't disagree with capture sharpening; I just didn't understand why it was there.  FWIW, I recommend deconvolution / SS for capture sharpening as I feel mixed-mode sharpenings are less inclined to artifact at output, but in the end it's all personal preference.

Mask Photo wrote:
I did lots of playing around to find a good value for my images, but during the write-up, I realized that the value would have to vary based on the size of the file. and then i did my math wrong. ;-\ I actually think I used 1/1000 of the longest image size as an arbitrary starting point that produced pleasing results.

Someone else does something not dissimilar on here.  Thanks for the explanation.

Koray wrote:
Sean - Have you ever thought of making a separations using negative clarity imports from conversion softwares?

Or maybe using their noise removal tools etc?

Actually I'll give them a try now

Yes, but that just becomes the time when I actually want to own C1.  And since I can't afford it, I avoid thinking about it .  Besides, I know what's coming in CS5 now and that (should) be incredible enough - at least for now .

Koray wrote:
whats coming in CS5?

Non-disclosure Agreement.

Elite Retouch wrote:

Non-disclosure Agreement.

Oh really?

Anyway I wont be around to use it at all...CS 6 maybe

Elite Retouch wrote:

Non-disclosure Agreement.

Not in my case - I wish!

Most of the features have been demo'd already online, but the one which hasn't been officially previewed and which referring to is this one.  Like when Photomatix was released, it will lead to a MM / Flickr / internet glut of images which rely on it as a one-shot wonder; but a few of the folks around here will figure out quite quickly how to really put its magic to work - for intended and unintended purposes.

shawnmakesfoto wrote:
Interesting, so you guys only split into high frequency and low frequency?

I split into anywhere from 5-8 levels of frequency.

holy cow. i can't fathom doing the same edits  2 or 3 or 5 times.
what benefit do you see from that extra work (err, precision)?

shawnmakesfoto wrote:
Interesting, so you guys only split into high frequency and low frequency?

I split into anywhere from 5-8 levels of frequency.

If you refer to the original thread, you'll see that many are using more than a single split; for some a single one is sufficient, depending on technique / intent.

Fotographia Fantastique wrote:
Thanks for posting this.
Before you had to find the "High Pass sucks + solution" thread, then search for the right posts.
This is much more concise.

Which posts do you want to see linked from the OP?   I've tried to keep the popular ones linked - did I miss some?

pellepiano wrote:
What I miss is some images that actually show the benefits of frequency splitting vs highpass and blur method.

For a thread that wish to simplify things, it would be great with some visiuals.

https://www.modelmayhem.com/po.php?thre … ost9585977

pellepiano wrote:
Thanks for the link. It does not make anything clearer to me though .

I was hoping for a AB comparison of a end result using both methods.

Final images side-by-side?  And if so, using the method for what purpose?

I'm happy to do what I can to make the principles more clear, but as I misunderstood your intent the first time...

pellepiano wrote:
Thanks for the link. It does not make anything clearer to me though .

I was hoping for a AB comparison of a end result using both methods.

in the end result depends on the retoucher. I used to use highpass instead of apply image to do the split (until Sean showed me the path to Cleanliness and Holiness) but I got through the lack of precision (especially on highlights and shadows) by masking or using regular healing on empty layer.

But I knew I was losing some information. Now I KNOW that information is less (I still have issues with the banding tho... looking into that )

Visual: Look up the highlights in the eyes in the HP and in the HF

x

Natalia_Taffarel wrote:
But I knew I was losing some information. Now I KNOW that information is less (I still have issues with the banding tho... looking into that )

Where are you getting banding?  Examples?

I have a night off; might as well put it to use...

SRB Photo wrote:

Where are you getting banding?  Examples?

I have a night off; might as well put it to use...

Everywhere, it's hardly noticeable but I NOTICE and it bugs me.

x

SRB Photo wrote:

Final images side-by-side?  And if so, using the method for what purpose?

I was looking for any purpose or example where the frequency separation really shines as opposed to the highpass method.

I will look into what Natalia said about the eyes.

pellepiano wrote:
I will look into what Natalia said about the eyes.

And then think... about all the information that you LOSE without seeing it

Black and white clipping point is "obvious" - other areas not so much...

On another note, the inverted highpass is a a lot more subtle working with edges when done with apply image and texture of the end result is a lot more natural (Because how much more detailed the apply image based highpass is)

I know this is still not visual and just Nataila's yara yara, but I'm hoping you'll take my word for it

x

Natalia_Taffarel wrote:

Everywhere, it's hardly noticeable but I NOTICE and it bugs me.

x

I guess I'm wondering if it's true banding or is just the PS preview engine showing false banding because of the approximations it uses?

Mask Photo wrote:

holy cow. i can't fathom doing the same edits  2 or 3 or 5 times.
what benefit do you see from that extra work (err, precision)?

It allows me to do stuff like shrink lines under eyelids, shrink pores, make changes to the larger base gradients without effecting smaller scale unevenness in skin tone, or vice versa.

It really makes things quicker, when you are working on very targeted layers, you have to be less precise, and simple healing brush work nearly always is sufficient.