Sean Baker wrote:

My preference would be for the layers in that order.  And if I might speculate, I think that grahamsz was suggesting a truly blank, normal layer.  One in which cloning / healing / real painting could occur.  But I've been wrong before - just ask my wife .

OK. I'll make a suggestion, though. I can place all but the original layer into a group and keep the original layer on the bottom. Then group visibility can be turned on and off to check the difference. The added benefit is it allows for changing opacity, fill, and blending mode of the group overall. Not that this is always useful or perhaps not ever useful for some, but it gives an added feature to the overall process if wanted.

If this seems useful to you, let me know. And I'm sure your wife would say you have never been wrong (while rolling her eyes, of course).

Photons 2 Pixels Images wrote:
For this Empty LF adjustment layer, are you talking a curves layer?

I'm also guessing this is the order you want them in from top to bottom.

I was thinking of a normal layer that I could paint onto to adjust the low frequency layer, so I can non destructively recolor things. I tried that on some work last night and it went well. When i was done painting onto the adjustment layer, i merged it with a copy of the original and reblurred to create a new low frequency layer that had the right frequency cutoff for my high frequency layer.

And yes, that was top to bottom - i keep a copy of the original on the very top so I can flip it on and compare the difference.

Robert Randall wrote:

When you've hung out in here long enough, you come to realize there are a large number of people that can't accomplish much in life. They need some form of reinforcement in their lives, and they seem to have chosen the art card for help with that. They can't make art, but they can hide behind the art card. Their arguments use the ploy of deflection to acheive credibility, much the same that OJ did during his murder trial. Everyone knows they are talentless hacks, but when they produce a snap shot of their girlfriend tugging at her soiled panties while sticking her ass out and giving the cell phone that come hither look that reminds you of a deer about to meet it's maker, they argue the picture is their art, and that somhow validates everything, and they are no longer a talentless hack screwing a toothless hag, they am be ARTEESTS!

Of course in this particular case, I'm sure the respondent is a highly talented artist that has a firm belief in the soul of his work. 

Or not.

*Stands and applause.

grahamsz wrote:

I was thinking of a normal layer that I could paint onto to adjust the low frequency layer, so I can non destructively recolor things. I tried that on some work last night and it went well. When i was done painting onto the adjustment layer, i merged it with a copy of the original and reblurred to create a new low frequency layer that had the right frequency cutoff for my high frequency layer.

And yes, that was top to bottom - i keep a copy of the original on the very top so I can flip it on and compare the difference.

OK. I got it basically the way you want it. I just put the created layers into a group that you can toggle the visibility on/off and see the changes just like if you put the original on top and toggle that layer off/on. It allows for a bit more flexibility here, though some may never use it and still gives you what you want. If you don't like it, let me know and I'll change it for you.

http://www.nunuvyer.biz/Photoshop/Frequency.zip

If you work in RGB/16 you won't need the script. If you work in any other mode, you will since the actions are based on RGB/16.

Photons 2 Pixels Images wrote:
http://www.nunuvyer.biz/Photoshop/Frequency.zip

Looks good to me

grahamsz wrote:
I think i've come up with something even more interesting. Here's a step by step tutorial on how I applied a selective bandstop filter to eliminate skin noise with periods between 5 and 15 px.

Here's the original image:


I then use photons 2 pixels action above to perform spatial frequency separation using a 5px blur. Once that's done I move to the low frequency layer and lasso the skin, taking care to avoid the eyes, lips and hair.


Now I apply a 15px feather, followed by a 15px gaussian blur. This in effect eliminates a particular range of frequencies in the skin, which in signal processing should make it bandstop filter.


And finally i switch to the high frequency layer and clone out blemishes


Total editing time for this little sample, about 3 mins.

Apologies again to the model here. She's actually got great skin, I just have a bunch of shots of her with fairly hash sunlight hitting her skin at an unflattering angle. I really should go take some of myself and use those.

That really evens out the tones across the skin, doesn't it? WOW. That's awesome.

Thanks.

grahamsz wrote:
I think i've come up with something even more interesting.

Ehehe.. I confirm, very interesting. If you keep going like this you will reduce my editing time a bit to much.

I tried this "new" approach on a full body, allready edited with D&B only. It was a swimsuit pola (in studio) with a lot of blotched skin (legs mostly). Anyway.. If you are careful enough when you apply the blur and combine it with good masking (no blurred-over-processed look).. Well.. I printed the two files and there is almost no difference (the fast one seems even better with very good texture on the skin).

Editing time for d&b: almost 3 hours
Editing time with Sean/Grahamsz method: almost 30 minutes

What else to say? Thank you!

I'm traveling atm, but when i come back at my office i will try to push it further (and post the result here).

a bit more to the left wrote:
Thanks for the posting and the actions scripts.

I'm not a pro but I've been using PS for quite awhile and I understand what's been said and still... I have a few questions.

First question...
I was always told that sharpening should normally be the very last edit you make to an image but everyone seems to be using the the technique as though it was something they would do after level or curve adjustments and before blemish retouching. Isn't sharpening really just increasing the contrast of the image?

I know that this isn't actually sharpening the image, it's just filtering it into two separate layers but you can't work on two separate layers at once as far as I know. Wouldn't it be best to completely edit the image first and then do this as a last step to accurately/creatively sharpen it?

I know that everyone has (been taught or developed) different methods for image editing but what I want to know is... at what stage should I actually be using this technique?

Second question...
Why Gaussian Blur? Why not Surface Blur or Lens Blur?

I tried both and combination of them all.
- Surface Blur produces a pre-sharpened/pre-blurred result that isn't good enough for print but would work for the web.
- Lens Blur seems to produce similar results to Gaussian Blur but it takes longer and if you use a mask with it, you can get similar results as Surface Blur.
- If you Surface Blur and then Gaussian Blur the Low Frequency layer you pretty much get the same result as just using Gaussian Blur.
- If you Surface Blur the Low Frequency Layer and then use a copy of that as the High Frequency Layer and then Gaussian Blur the Low Frequency Layer you get an image where the blemishes aren't sharpened.

I know you're looking for accuracy in the filter but that accuracy is only useful by itself. As soon as you have created the high and low frequency layers you have to edit them for them to be of any use... so why the need for accuracy?

I can't answer all this but most people perform capture sharpening on an image initially. Most digital cameras have an anti-aliasing filter which causes slight blur so a bit of initial sharpening will get you back to the original image....so to speak. Then most will sharpen again for output......depending on where the output is going: web, print, etc.

I can't really talk about the difference between the filters. I'll let those who know more about it answer.

JeF Briguet wrote:

Ehehe.. I confirm, very interesting. If you keep going like this you will reduce my editing time a bit to much.

I tried this "new" approach on a full body, allready edited with D&B only. It was a swimsuit pola (in studio) with a lot of blotched skin (legs mostly). Anyway.. If you are careful enough when you apply the blur and combine it with good masking (no blurred-over-processed look).. Well.. I printed the two files and there is almost no difference (the fast one seems even better with very good texture on the skin).

Editing time for d&b: almost 3 hours
Editing time with Sean/Grahamsz method: almost 30 minutes

What else to say? Thank you!

I'm traveling atm, but when i come back at my office i will try to push it further (and post the result here).

I think I can figure out a way to include this latest step from grahamsz in with the actions. I'll work on that today.

a bit more to the left wrote:
Thanks for the posting and the actions scripts.

I'm not a pro but I've been using PS for quite awhile and I understand what's been said and still... I have a few questions.

First question...
I was always told that sharpening should normally be the very last edit you make to an image but everyone seems to be using the the technique as though it was something they would do after level or curve adjustments and before blemish retouching. Isn't sharpening really just increasing the contrast of the image?

I know that this isn't actually sharpening the image, it's just filtering it into two separate layers but you can't work on two separate layers at once as far as I know. Wouldn't it be best to completely edit the image first and then do this as a last step to accurately/creatively sharpen it?

I know that everyone has (been taught or developed) different methods for image editing but what I want to know is... at what stage should I actually be using this technique?

Second question...
Why Gaussian Blur? Why not Surface Blur or Lens Blur?

I tried both and combination of them all.
- Surface Blur produces a pre-sharpened/pre-blurred result that isn't good enough for print but would work for the web.
- Lens Blur seems to produce similar results to Gaussian Blur but it takes longer and if you use a mask with it, you can get similar results as Surface Blur.
- If you Surface Blur and then Gaussian Blur the Low Frequency layer you pretty much get the same result as just using Gaussian Blur.
- If you Surface Blur the Low Frequency Layer and then use a copy of that as the High Frequency Layer and then Gaussian Blur the Low Frequency Layer you get an image where the blemishes aren't sharpened.

I know you're looking for accuracy in the filter but that accuracy is only useful by itself. As soon as you have created the high and low frequency layers you have to edit them for them to be of any use... so why the need for accuracy?

OK. I reread this post and I think I can answer some of it. Once the layers are separated, your low frequency layer holds the information for the larger areas between detail. This layer is best suited for evening out the tones or removing larger blemishes. The high frequency layer is best suited for removing spot blemishes, hair strands, smaller stuff without effecting the skin tones much. Some things will have to be taken care of on both layers. But, this does drastically cut down on time...at least it has for me so far. I find I have to resample a lot less with the healing brush to get good results.

I've just started using this method so I'm still experimenting, but this is what I've found so far. I'm now setting this up as my initial step and using the healing brush to edit the skin. Then I go on to the capture sharpening.

Photons 2 Pixels Images wrote:
Anyone want this step included in the action set?

The way i used it on my test was like this:

Copy the low frequency layer..

Select a part where the skin is uneven.. Feather, apply blur (sometimes less than 10px, somtimes more), mask it (wacom with low opacity).

And then.. Select another part, etc..

So a separate action is best for my workflow.

grahamsz wrote:
I think i've come up with something even more interesting. Here's a step by step tutorial on how I applied a selective bandstop filter to eliminate skin noise with periods between 5 and 15 px.

A wonderful technique, grahamsz.  Are you sure you're just starting to edit?   If I may, I'd like to add a few notes:
  1.) This technique will result in a very small loss of tonal contrast across the applied area.  Think back to grahamsz' description of component waves making up a whole 'image' and remember that while most of your 'problem' waves lie in the bandpass zone, there are some 'good' waves in there too contributing to the overall contrast across the selection.  Smart Sharpen set to Gaussian Blur can help to mitigate this.  One can also use Surface Blur in place of the second GB, as it will lose less of the overall contrast.
  2.) Depending on the interaction of your original file dimensions, output dimensions, and frequencies eliminated, you may have difficulty with the skin looking 'plasticy' when resized (if you're really geeky, you can think of this entirely in terms of spatial frequencies and destructive interference).  Try adding texture back in with noise + embossing or the texturizer (try looking up Chip Springer's skin action for more on this method) and then band-passing that to include only what was taken out of the original image.  All skin should have some amount of texture at all frequencies in order to look natural at all viewing distances.
  3.) This technique lends itself wonderfully to models with good -> great skin; it has some difficulty if you shoot a woman (or man) with poor skin uncorrected by makeup.  Yes, I'm saying this from experience .  It's in these circumstances which I think grahamsz original post on skin correction, or of course using RNT's method, will give you the best results.  YMMV.

a bit more to the left wrote:
I know you're looking for accuracy in the filter but that accuracy is only useful by itself. As soon as you have created the high and low frequency layers you have to edit them for them to be of any use... so why the need for accuracy?

For the same reason I like to know how to separate the red, green, and blue channels accurately - sometimes the data have value independent of one another.  If I'm doing skin correction, there certainly is some validity to saying that accuracy isn't of the utmost importance.  For example, the skin technique is traditionally described as a high pass layer which has a gaussian blur run on it at 1/3 the HP value, inverted, and applied at 50% linear light - and for most purposes, that's good enough.  But when sharpening or creating selections, artifacts induced by a less-accurate methodology become much more noticeable.  Inherently I'm increasing contrast in both operations, and if I do so inaccurately it will appear out of place in the image.

I hope that made sense.  If not, say so - I'll drink more coffee and repeat.

Sean

Sean Baker wrote:
A bit OT, but is anyone else having difficulty opening P2P's action + script combo?  I'm sure I'm doing something entirely stupid, but it won't let me extract the script portion on Vista 64.

What's it doing to you?

I used an evaluation version of Winzip to make it and used the old zip method for backward compatibility.

Sean Baker wrote:
A wonderful technique, grahamsz.  Are you sure you're just starting to edit?

Yeah, i've been editing people for about a week. It didn't take me long to get frustrated and need to find a better approach.

Sean Baker wrote:
1.) This technique will result in a very small loss of tonal contrast across the applied area.  Think back to grahamsz' description of component waves making up a whole 'image' and remember that while most of your 'problem' waves lie in the bandpass zone, there are some 'good' waves in there too contributing to the overall contrast across the selection.  Smart Sharpen set to Gaussian Blur can help to mitigate this.  One can also use Surface Blur in place of the second GB, as it will lose less of the overall contrast.

Actually, where i've used this in practice, I've blended some of the original low pass layer in on the face and only left it out altogether where there are blotches on the legs. Photons 2 Pixels idea of painting on the secondary blur seems like a winner.

Sean Baker wrote:
2.) Depending on the interaction of your original file dimensions, output dimensions, and frequencies eliminated, you may have difficulty with the skin looking 'plasticy' when resized (if you're really geeky, you can think of this entirely in terms of spatial frequencies and destructive interference).  Try adding texture back in with noise + embossing or the texturizer (try looking up Chip Springer's skin action for more on this method) and then band-passing that to include only what was taken out of the original image.  All skin should have some amount of texture at all frequencies in order to look natural at all viewing distances.

I'll look into that.

I should of course point out that my example included a couple of pixel choices that will depend a lot on your sensor size and how close you are to the model. Nothing about these numbers is anything more than trial and error.

Sean Baker wrote:
3.) This technique lends itself wonderfully to models with good -> great skin; it has some difficulty if you shoot a woman (or man) with poor skin uncorrected by makeup.  Yes, I'm saying this from experience .  It's in these circumstances which I think grahamsz original post on skin correction, or of course using RNT's method, will give you the best results.  YMMV.

There's certainly no magic solution here, but I think we've got some pretty powerful tools here to apply to most problems. I retouched a model with some sunburn yesterday and used a selective color adjustment on the low frequency layer to remove the redness then hand blended that in until it looked smooth.

grahamsz wrote:
¨¨
Actually, where i've used this in practice, I've blended some of the original low pass layer in on the face and only left it out altogether where there are blotches on the legs. Photons 2 Pixels idea of painting on the secondary blur seems like a winner.

Same here. I won't use it on a face but it works very well on blotched limbs.

If you are subtle and mask out you don't see that much difference in the output (worked on a MF file). And well.. It's very fast compared to my usual d&b routine.

Now for closeups, even with a very good makeup, i won't even try.

Photons 2 Pixels Images wrote:
OK. I repacked it and reuploaded it. Try it again and see if it works. I used a different zip utility this time.

Not sure what the difference was, but it's working now - thank you.  A brilliant bit of code with sooo many uses.

grahamsz wrote:
There's certainly no magic solution here, but I think we've got some pretty powerful tools here to apply to most problems. I retouched a model with some sunburn yesterday and used a selective color adjustment on the low frequency layer to remove the redness then hand blended that in until it looked smooth.

Agreed.  And please don't think I'm trying to belittle the techniques you're using - whether I'm doing it well or not, I'm trying to help you expand them.

grahamsz wrote:
Yeah, i've been editing people for about a week. It didn't take me long to get frustrated and need to find a better approach.

I suppose in the interests of disclosure, I should add that I did a CS&E major so i've done college level computer graphics and signal processing courses.

When i first got my hands on Photoshop 1.0 my reaction was to the tune of "this is cool, but it's going to be a really long time until monitors, scanners and cpus make this as useful as it could be". I think we might be there now

Sean Baker wrote:
Agreed.  And please don't think I'm trying to belittle the techniques you're using - whether I'm doing it well or not, I'm trying to help you expand them.

Not at all. I'm a big fan of collaboration and learning from others :-)

grahamsz wrote:

Not at all. I'm a big fan of collaboration and learning from others :-)

Same here. I'll explain anything I know to anyone who will listen. Granted, on the Photoshop front my knowledge is very limited, but when it comes to automating tasks in almost any aspect of computers, I can usually find a way.

I'm actually now going back through some of my older stuff that I thought was unusable and trying this technique on them with surprising results.

If there is anything else you want added to this script/action set let me know. I can probably make it customizable so everyone can have it exactly the way they want it. This is so useful for so many things, it would be worth the effort.

Photons 2 Pixels Images wrote:
I'm actually now going back through some of my older stuff that I thought was unusable and trying this technique on them with surprising results.

I'm doing the same exact thing... I may end up making some actions myself in the next few days, we all have out own particular way of doing things.

Robert Randall wrote:
I've been doing imaging work since 1989, on more platforms than I can count, and I've not run into anything like this before. You have no idea how exciting it is to see something really new to me. Just wanted you guys to know how much I appreciate what you are doing, and please keep experimenting.

Experimenting with things like this is one of the few things that brings joy to my life. I've gone off on a tangent now with this trying to break down the math a bit more for a better starting point with the radii settings.

Question for all who have been trying this: Do you find that the radius for the GB on the low frequency layer is pretty standard or do you often change it based on the image?

If you change it like I've been doing, what criteria do you go by? Just how it looks? Until it hits a certain level then maybe back off a bit? Does it seem to be based on size?

Some extra explanation on that last one...if you are shooting a closeup headshot does it normally work out to where you set the radius higher or lower than if you shot one where the subject was further away? What I'm thinking is that if we were to draw a selection around the head of the subject (simple rectangle) would the resolution of that selection compared to the radius in the GB be fairly consistent?

I hope that all makes sense. If anyone wants to just keep this in mind while doing this and let me know if there seems to be a pattern, I'd appreciate it.

Photons 2 Pixels Images wrote:
Question for all who have been trying this: Do you find that the radius for the GB on the low frequency layer is pretty standard or do you often change it based on the image?

Extremely variable here.

Photons 2 Pixels Images wrote:
If you change it like I've been doing, what criteria do you go by? Just how it looks? Until it hits a certain level then maybe back off a bit? Does it seem to be based on size?

Appearance mostly, though you can sort of guess what the radius will be based on image dimensions, field of view, and % of the composition which the skin represents.

Photons 2 Pixels Images wrote:
Some extra explanation on that last one...if you are shooting a closeup headshot does it normally work out to where you set the radius higher or lower than if you shot one where the subject was further away? What I'm thinking is that if we were to draw a selection around the head of the subject (simple rectangle) would the resolution of that selection compared to the radius in the GB be fairly consistent?

So long as you're always selecting the same 'size' (actual life size) area, then yes I believe it would be.  There will be some variation based on the individual's skin and the types of deformities present as well as with the intended final look, but I would agree that this is on the right track as far as defining a starting point to work from.

Sean Baker wrote:
So long as you're always selecting the same 'size' (actual life size) area, then yes I believe it would be.  There will be some variation based on the individual's skin and the types of deformities present as well as with the intended final look, but I would agree that this is on the right track as far as defining a starting point to work from.

This is what I've noticed and it seems to be really close each time based on resolution of the head area of the subject. I'm not even sure this will be worth it, but I want to see if I can set a baseline then adjust for the relative size/amount of imperfections.

I'm not even sure this will be worth it, but if it is it will speed up workflow just a bit more. If it can save even 1 minute for each image, that's enough for me when I have 100+ images to try to get through.

Photons 2 Pixels Images wrote:
OK. Here's what I got. Let me know what you guys think.....

1. Checks mode and bit depth of image. Saves original settings of these.
2. Converts to RGB mode and 16 bit.
3. Flattens image to single layer if not already.
4. Creates Low Frequency layer with popup GB dialog, default at 2.0 pixel radius.
5. Creates High Frequency layer using the subtract method.
6. Creates a Dodge/Burn layer above these with neutral gray in Overlay blend mode.
7. Puts those 3 layers into a group above the original layer with Pass Through blend mode.
8. If there was a conversion in step 2, it converts back to the original mode and/or bit depth.

This uses a combination script and action set. The only thing the script really does is checks modes, converts if necessary, then converts back at the end. Basically, the steps in italics above. If you normally work in RGB/16 you don't need the script.

I was playing with the action a little bit, the thing I personally don't like is #3, I never flatten my PSD files. I'd rather see all layers merged into a new layer on top and the script run on that. In my workflow, prior to sharpening I do the Shift-Ctrl-Atl-E dance on the keyboard and make myself a flattened layer that I do any sharpening to.

Robert Randall wrote:
I've been doing imaging work since 1989, on more platforms than I can count, and I've not run into anything like this before. You have no idea how exciting it is to see something really new to me. Just wanted you guys to know how much I appreciate what you are doing, and please keep experimenting.

Sean, Grahamsz, and Photons 2 ....  Mr. Randall doesn't make comments like this for just anyone.  All of you should be very proud of what you've posted and shared.  Special thanks to Sean for starting this thread.  It's the best damn thing I've seen on MM in a very long time!!!  Truely amazing!!!

ACPhotography wrote:

I was playing with the action a little bit, the thing I personally don't like is #3, I never flatten my PSD files. I'd rather see all layers merged into a new layer on top and the script run on that. In my workflow, prior to sharpening I do the Shift-Ctrl-Atl-E dance on the keyboard and make myself a flattened layer that I do any sharpening to.

I can make a change to that. I'll let you know once it's updated. Thanks.

KEKnight wrote:

Sean, Grahamsz, and Photons 2 ....  Mr. Randall doesn't make comments like this for just anyone.  All of you should be very proud of what you've posted and shared.  Special thanks to Sean for starting this thread.  It's the best damn thing I've seen on MM in a very long time!!!  Truely amazing!!!

grahamsz wrote:
Thanks for all the great responses and thanks to Sean for kicking this whole thing off.

I'm pretty excited about this and thrilled that so many accomplished photographers feel the same way.

Sean deserves the main kudos for this. And grahamsz. All I did was put it all together into a script/action to make it easier.

I also think it's turned into more of a community effort. I just hope people keep posting here to let us all know of any more uses you've found for this or anything else you've tried to go along with it to make it better.

So, thank you Sean and grahamsz....and thank you to everyone else who has contributed.

ACPhotography wrote:
I was playing with the action a little bit, the thing I personally don't like is #3, I never flatten my PSD files. I'd rather see all layers merged into a new layer on top and the script run on that. In my workflow, prior to sharpening I do the Shift-Ctrl-Atl-E dance on the keyboard and make myself a flattened layer that I do any sharpening to.

OK. New action set up that does what you're asking. At least I hope it does.

http://www.nunuvyer.biz/Photoshop/Frequency2.zip

This one just has the actions...the script is the same.

This action set is for RGB/16. If you work in another mode, please get the script/action set here and use the script contained within the zip file:

http://www.nunuvyer.biz/Photoshop/Frequency.zip