Damien Menard wrote:
Thanks so much for the seperation walkthrough! I played with some of the bandstop-type techniques and didn't like the look, but just being able to work on the different freq's using the usual tools has made things so much easier.

I tried to repeat Sean's process on the high-frequency layer in order to further seperate the image, but didn't have much luck. It seems like blurring the high-freq layer just results in a blurry highpass-looking thing, instead of reverting it back toward the unseperated image.

Would anyone be able to post a walkthrough similar to Sean's for further seperating the image after going through Sean's technique?

Up above on the previous page I posted a link to a "Sharpen.jsx" script for Photoshop that will do further separations. It currently does 3 separations giving you 3 High Frequency layers each clipped with a curves adjustment layer. You can then adjust the curves for more or less contrast on that frequency as well as adjusting opacity.

I don't have much time now, but I can maybe later give a rundown of what it does.

Photons 2 Pixels Images wrote:

I don't have much time now, but I can maybe later give a rundown of what it does.

Thanks, I appreciate it.  Running curves on individual frequencies is a cool idea, but right now I'm looking for something that'll keep the combined layers looking as close to the untouched background layer as possible.

Damien Menard wrote:

Thanks, I appreciate it.  Running curves on individual frequencies is a cool idea, but right now I'm looking for something that'll keep the combined layers looking as close to the untouched background layer as possible.

If you adjust the curves to lower the contrast (inverse "S" curves) along with the opacity, you can achieve this. The default is set to oversharpen. I am working on a method of determining the exact settings to get an exact replica of the Background, but so far it seems to be elusive.

As for further separating the High Frequency layer, I don't think it will work since it's a 50% neutral gray layer type.

If I figure anything else out, I'll let you know.

Sean Baker wrote:
Findings / Technique: In my own experimentation, I've found that HP gives differences as high as 2670/32k per pixel when separating high and low frequency information.

what is your exact method to separate HF and LF with HP?
in 8bit i found HP to give slightly better results in some areas, like out-of-focus fine hair detail, at least to my eyes
example (layered tif crop,

PANZERWOLF wrote:
what is your exact method to separate HF and LF with HP?
in 8bit i found HP to give slightly better results in some areas, like out-of-focus fine hair detail, at least to my eyes
example (layered tif crop,

Photons 2 Pixels Images wrote:

If you adjust the curves to lower the contrast (inverse "S" curves) along with the opacity, you can achieve this. The default is set to oversharpen. I am working on a method of determining the exact settings to get an exact replica of the Background, but so far it seems to be elusive.

As for further separating the High Frequency layer, I don't think it will work since it's a 50% neutral gray layer type.

If I figure anything else out, I'll let you know.

I'm not sure if I'm on the same sheet as you guys are discussing here, but it's possible to create near infinite levels of separation without meaningful loss (albeit loss does become additive per iteration).  Here is a multi-layer TIFF (~2.8MB) containing 6 orders of frequency separation with a cumulative loss of roughly (6,5,4) [R,G,B] (16bit) in the worst areas.  This done by duplicating and re-blurring the 'LF' layer repeatedly.  Though the provided file is in 8bit for size purposes, all separation was done in 16bit for accuracy.  Hope that helps.

Sean Baker wrote:

I'm not sure if I'm on the same sheet as you guys are discussing here, but it's possible to create near infinite levels of separation without meaningful loss (albeit loss does become additive per iteration).  Here is a multi-layer TIFF (~2.8MB) containing 6 orders of frequency separation with a cumulative loss of roughly (6,5,4) [R,G,B] (16bit) in the worst areas.  This done by duplicating and re-blurring the 'LF' layer repeatedly.  Though the provided file is in 8bit for size purposes, all separation was done in 16bit for accuracy.  Hope that helps.

Yes, you're on the same page. That's basically what the latest script I posted does. It gives a GB dialog for the first separation, then automatically separates at 1/2 that radius for the second, then 1/4 radius for the third and stacks the HF layers onto the last GB layer.

I think what he's saying, and what I've noticed, is when you do this you will see an additive effect on the HF layers so it won't look the same as the original. I've noticed the same thing, but usually will adjust the curves and/or opacities of the HF layers to bring it back to the same appearance. This allows me to filter out specific frequencies while still retaining some of the detail.

I've also noticed that the appearance differs depending on the stacking order. If you stack top to bottom HF1, HF2, HF3 it looks different than if you stack them in reverse order.

I'm trying to figure out if there is a mathematical relationship between the opacities of the HF layers and how they are stacked that will result in an image appearance the same before and after separation regardless of how many separations are used.

Photons 2 Pixels Images wrote:

Yes, you're on the same page. That's basically what the latest script I posted does. It gives a GB dialog for the first separation, then automatically separates at 1/2 that radius for the second, then 1/4 radius for the third and stacks the HF layers onto the last GB layer.

I think what he's saying, and what I've noticed, is when you do this you will see an additive effect on the HF layers so it won't look the same as the original. I've noticed the same thing, but usually will adjust the curves and/or opacities of the HF layers to bring it back to the same appearance. This allows me to filter out specific frequencies while still retaining some of the detail.

I've also noticed that the appearance differs depending on the stacking order. If you stack top to bottom HF1, HF2, HF3 it looks different than if you stack them in reverse order.

I'm trying to figure out if there is a mathematical relationship between the opacities of the HF layers and how they are stacked that will result in an image appearance the same before and after separation regardless of how many separations are used.

My gestalt is that the issue with the script is coming from the fact that each separation isn't being done from the same base image, as well that the LF layer isn't retained under it, making perfect recreation of the original impossible.  The unfortunate truth of GB is that it isn't a perfect frequency separation, but only an approximation.  Hence, running a 4px GB on itself repeatedly will continue to change the base image.  Equally, running multiply on separate copies will not only result in keeping extra copies of the very fine details, but will also result in different 'versions' or 'interpretations' of the intermediate frequencies.  The mathematical issue with layer ordering I suspect has to do with the way we minimize the fine details' numerical values in the technique, leading to their being averaged away by the larger structures in recombination.  Frustrating to be sure, but I'm not sure there's a better way with the available blend modes and their implementation - maybe something else to hope for in CS5?

Brian T Rickey wrote:
I am wondering of there is a paper saving way of printing this thread?

I'd go through and just pull out the posts which you find most valuable, though I'm sure there are some print wizzes with better ideas than that.  And I'm equally curious to read them .

Sean Baker wrote:
Looking at the file you posted up (love that you gave an example btw) it looks like you used the 16bit technique for the method outlined in this thread on an 8bit image, thus imparting the slight difference from the original to the result.

no, i used your 8bit method (in 8bit mode) and compared it to my HP method

Sean Baker wrote:
OK, technique when using the actual HP filter should be: Create two copies of the image.  Run HP on the top copy; set blend to Overlay+100% or LL+50%.  Run GB on the bottom copy; blend mode Normal+100%.  The problem is that the image doesn't look the same as it did before, illustrated in the example posted here.

oh, i see
but setting the LL layer to 50% opacity is not the same as reducing the contrast of the layer to 1/2!
both overlay/100 and LL/50 give truly horrible results (resulting in your high difference), but that's not what i'm doing

in my tif file you can see the HP layer (just HP, no alterations) on LL, 100% opacity,
and on top of it the curves layer that lowers the contrast
in 8bit, this is (at least) as accurate as your method
in 16bit, i find yours to be a tad more accurate in most cases, but i guess for both these are mere rounding errors, not by far the extent shown in your difference example

am i correct that your initial HP bashing and he figure 2670/32k was based on an incorrect HP method, or am i still missing something?

PANZERWOLF wrote:
no, i used your 8bit method (in 8bit mode) and compared it to my HP method

Gotcha.

PANZERWOLF wrote:
oh, i see
but setting the LL layer to 50% opacity is not the same as reducing the contrast of the layer to 1/2!
both overlay/100 and LL/50 give truly horrible results (resulting in your high difference), but that's not what i'm doing

This is an interesting and important point which I don't recall having been elucidated previously (my apologies to the poster if it has) - the math by which both blend modes work does in fact impart a difference at the high and low ends when using modified opacity settings.  Of course, the way I look at it, it's just more reason to not use the HP filter (see below).

PANZERWOLF wrote:
in my tif file you can see the HP layer (just HP, no alterations) on LL, 100% opacity,
and on top of it the curves layer that lowers the contrast
in 8bit, this is (at least) as accurate as your method
in 16bit, i find yours to be a tad more accurate in most cases, but i guess for both these are mere rounding errors, not by far the extent shown in your difference example

This appears to be due in large part to the example and radii which you are employing.  Differences are very small until you really push the radii, at which time larger ones become apparent.  Using your methodology, I had considerably more issue with difference from the original, particularly in extremely bright / dark areas of the fine detail.  I can post example files if you'd like to see them - HP + GB + Curves gave me as much as 75/256 difference; the 8bit technique at the outset of this thread showed 1/256 maximum, admittedly above what is posted there (I'll go back to edit).

PANZERWOLF wrote:
am i correct that your initial HP bashing and he figure 2670/32k was based on an incorrect HP method, or am i still missing something?

2670/32k was on the basis of a curve-corrected HP filter utilization IIRC, but again all differences are heavily impacted by radii employed - I'd have gotten much more or less depending on the radius I'd chosen.  One of the advantages of the techniques outlined is that their maximum 'damage' to a file is independent of the radius used.

Sean Baker wrote:
This appears to be due in large part to the example and radii which you are employing.  Differences are very small until you really push the radii, at which time larger ones become apparent.  Using your methodology, I had considerably more issue with difference from the original, particularly in extremely bright / dark areas of the fine detail.

ahh, i was trying radii from 1-4 only, i'll have to go further ; )
and regarding the extremely bright / dark areas, i guess the HP filter just overshoots on high contrast edges, with the resulting layer blowing highlights and blacks, which, of course, cannot be retrieved with curves, while the subtraction avoids this in the first place, since the scale brings it down to 1/2 during the computing

but i think i just found a solution: reducing the contrast of the original layer before applying the HP filter (not on the gauss layer of course) brings the same 1/2 contrast HP result, but without the overshooting damage

i just tried it on a harsh black/white grid with radius 20, my 1st HP method produced horrible edge/gap shadows (surely the same thing that you experienced), but the 2nd worked pretty well!

could you please post an example where you were getting problems with the HP method? the original layer for me to experiment on? would be cool if we worked on the same file ; )
a crop of a problem area is enough of course

damn, i like this thread = )

PANZERWOLF wrote:
ahh, i was trying radii from 1-4 only, i'll have to go further ; )
and regarding the extremely bright / dark areas, i guess the HP filter just overshoots on high contrast edges, with the resulting layer blowing highlights and blacks, which, of course, cannot be retrieved with curves, while the subtraction avoids this in the first place, since the scale brings it down to 1/2 during the computing

but i think i just found a solution: reducing the contrast of the original layer before applying the HP filter (not on the gauss layer of course) brings the same 1/2 contrast HP result, but without the overshooting damage

i just tried it on a harsh black/white grid with radius 20, my 1st HP method produced horrible edge/gap shadows (surely the same thing that you experienced), but the 2nd worked pretty well!

could you please post an example where you were getting problems with the HP method? the original layer for me to experiment on? would be cool if we worked on the same file ; )
a crop of a problem area is enough of course

damn, i like this thread = )

Heading to bed, but take a look at this as an example file (~2.8MB).  I don't think using just a small crop out of an image is going to give the best comparative result, simply for the need to use large radii to push the techniques.  Will post more tomorrow if you can take a look at it and make sure I used your method correctly.

As to HP operation, see the other post I made a bit above about that; I really think your comments on subtraction and curves were on the money, as it perfectly describes the filter's (aberrant) result.

EDIT: Looking even closer, I'm going to refine my statements to effect that you're right about base contrast being key to this and / or a way of getting around the HP limitations.  Indeed the 'correct' HP results are >128 levels wide, explaining the clipping observed both with Subtract (Scale 1) and the HP filter.

Mask Photo wrote:
I'm currently wrestling with a few things considering this workflow:
1) there doesn't seem to be a chance to run noiseninja. Are we talking about using the HF layer(s) for all noise elimination, or should I run it prior to splitting the layers?
2) I hear people talking about using it for sharpening; are they just duplicating the HF layer and adjusting the opacity, or is there something that I'm missing?
3) if sharpening is done by duplicating HF layers, then would an effective method for creative sharpening be to just generate a "throwaway" HF layer after all editing is done, put it on top of everything, and mask it to taste?

As to healing - using a separate layer can work, but I don't think it's going to do so well if the blend mode for the HF data is anything but Normal.  I'd be happy to be wrong about that (and may well be), but iirc, it won't quite grasp what you're trying to do.
Numerically,
1.) NN can be run on a separate copy and the noise removed to another layer just as is done for the HF / LF data - it's the same type of procedure and has the same 'lossless' nature to it.
2.) Generally, yes.  Some just separate and then run an S curve against the HF; others use it in duplicate.
3.) I think masking is a good idea, at least for creative sharpening, but not everyone will agree with this as it's all a matter of taste at that point.

Sean Baker wrote:
As to healing - using a separate layer can work, but I don't think it's going to do so well if the blend mode for the HF data is anything but Normal.  I'd be happy to be wrong about that (and may well be), but iirc, it won't quite grasp what you're trying to do.

It seems to work pretty well if the HF layer, the edit layer, and any clipped curves are all in a layer group with its blend mode set to linear light; observe my partial benjamin button treatment on my dad:
http://maskphoto.com/files/hf_test.psd (18MB)
(just horsing around with this file; there's no way i'd erase his rich lines of character for a portfolio piece)
The trick is to turn OFF the blending mode while healing, so you can't really see the effects (but you can see what the texture looks like still). You could also leave it on, and target a portion of "good" skin on the HF layer with which to paint on the HF retouch layer, but it would be a pain to retarget another area if you needed to change texture.

1.) NN can be run on a separate copy and the noise removed to another layer just as is done for the HF / LF data - it's the same type of procedure and has the same 'lossless' nature to it.

Ah, so if I understand you correctly, "apply image" can be used to split any kind of before/after procedure into two layers? So the procedure would be to run NN on one layer, then apply a duplicate layer to it, from which we'll get a layer with all the noise removed, and a delta layer with the noise put back in? From that point, masking the unwanted noise out is the only remaining step?

2.) Generally, yes.  Some just separate and then run an S curve against the HF; others use it in duplicate.

Is there a blend mode during apply image that would tend to amplify the effects, or is linear light the only way to go about this procedure?


Continuing on, I've read a little about a double-blur method. is that just to smooth out unevenness? and if so, wouldn't that cause problems in areas that aren't skin? I realize this thread has been dominated by skin-related discussions, but I've also seen coastlines and other non-people samples.
Was the secondary blurring the gist of the "bandstop" method mentioned by someone elsewhere, or was that something else?
(one would think that having a mathematics degree would grant me some insight into some of these topics, but that would be incorrect)

Thanks so much for your time on this method.

Mask Photo wrote:
Ah, so if I understand you correctly, "apply image" can be used to split any kind of before/after procedure into two layers? So the procedure would be to run NN on one layer, then apply a duplicate layer to it, from which we'll get a layer with all the noise removed, and a delta layer with the noise put back in? From that point, masking the unwanted noise out is the only remaining step?

Correct, though one can do a lot of other things with the principles involved as well - sharpening without noise enhancement as one among them - particularly useful for 'capture' sharpening w/ deconvolution.

Mask Photo wrote:
Is there a blend mode during apply image that would tend to amplify the effects, or is linear light the only way to go about this procedure?

You can play with it and you'll find that there are ways to amplify in a single step, but that's a bit beyond the scope of the thread - the original intent here was recreation of the image in multiple parts without losing any data.  Any / all of the amplification methods results in loss of fidelity with the original.

Mask Photo wrote:
Continuing on, I've read a little about a double-blur method. is that just to smooth out unevenness? and if so, wouldn't that cause problems in areas that aren't skin? I realize this thread has been dominated by skin-related discussions, but I've also seen coastlines and other non-people samples.
Was the secondary blurring the gist of the "bandstop" method mentioned by someone elsewhere, or was that something else?
(one would think that having a mathematics degree would grant me some insight into some of these topics, but that would be incorrect)

The double-blurring is almost always being applied to skin regions, either through manual selection or through masking of the result.  But yes, that is the basis of the visual bandstop which has been discussed.

Sean Baker wrote:

My gestalt is that the issue with the script is coming from the fact that each separation isn't being done from the same base image, as well that the LF layer isn't retained under it, making perfect recreation of the original impossible.  The unfortunate truth of GB is that it isn't a perfect frequency separation, but only an approximation.  Hence, running a 4px GB on itself repeatedly will continue to change the base image.  Equally, running multiply on separate copies will not only result in keeping extra copies of the very fine details, but will also result in different 'versions' or 'interpretations' of the intermediate frequencies.  The mathematical issue with layer ordering I suspect has to do with the way we minimize the fine details' numerical values in the technique, leading to their being averaged away by the larger structures in recombination.  Frustrating to be sure, but I'm not sure there's a better way with the available blend modes and their implementation - maybe something else to hope for in CS5?

I'm finding the same thing through experimentation of different ways to do this. I've tried separating from a GB of the same base image at different radii and additive GB to the same base. Each method has its pros and cons. And I think you hit on the main drawback to multiple separations and that being the available blend modes. This is what led me to try the clipped curves and opacity adjustments.

This just gives me more to play with. However, I'm definitely not knocking the potential that this separation method has nor the real world uses as it is. And I'm having fun figuring out what else I can do with it.

Sean Baker wrote:
On real-time skin work:

Spawning off of what Panzerwolf started above, I've gone back to look more closely at what's going on with the High Pass filter proper and how we might exploit the new insights into its operation.  As stated, its method of applying the change can't handle a full-contrast scene as the range involved is sometimes too great for it to record.  I was hesitant to use PW's Curves techniques as they were imperfect for halving the contrast and recreating the scene perfectly (or at least to the standard previously set), despite countless attempts at changing curve values and playing with all manner of permutations, opacities, etc.  It should be noted that I don't believe this is the result of his efforts, but of limitations inherent to PS itself.  Curves didn't work; blending with 50% gray didn't work; blending with shifted grays didn't work (creating a 127.5 gray); but the Legacy Brightness / Contrast adjustment works. 

I know, I'm shaking my head over that one too, but let's use it:

1.) Take your image on which you want to do some skin work, creating a copy on which you'd like to work.
2.) Use Image->Adjustments->Brightness / Contrast; check the Legacy box; enter -50 in the Contrast box; 0 Brightness.
3.) Invert the layer.
4.) Convert the layer to a Smart Object.
5.) Run High Pass first, selecting your upper pixel limit for elimination.
6.) Run GB next, selecting the lower limit of elimination / upper limit of preservation in the detail.
7.) Tune to taste, rasterizing if you so choose for speed / size.

That Brightness / Contrast gives us a better result than any Curve or gray blend raises a lot of questions, but those will have to wait for another day.  In the mean time, you can do realtime skin bandstopping without loss of pixel-level quality (yes Virginia, you do retain 1/32k accuracy) .

Where's the action lol

Julian Marsalis wrote:

Where's the action lol

On the right side of your screen, in the actions palette.  It's the round button .

Ken Marcus Studios wrote:
How do I know what the upper pixel limit is ?   any recommendation to start with?

What is GB and what is the lower limit of elimination/uper limit ?

I'm sure your technique is good . . . but you don't make it very clear as to how to do it


KM

The post is really meant as a continuation / derivative of a technique first mentioned in the thread by grahamsz here, wherein the premise is to eliminate a certain frequency range of image information corresponding to problem areas in skin (or whatever you happen to be editing).  That I didn't make that more clear was a mistake - sorry about that.

The upper pixel limit referenced will depend entirely on the relative area of skin in the image, as well the image's dimensions.  For high-MP headshots, this might be in the range of 25-50+, but again will depend on the issues to be overcome.

GB = Gaussian Blur.  The lower limit would be the smallest-sized detail which you would want to remain in the image - generally around 1.5-3x the diameter of a pore in the image, though this will again vary by taste and the image used.  The real value here is that you can play with these values as you go to see the impact they have on the base image.

ESCALANTE wrote:
I'm not sure if it has been asked yet or not but,
How is it effecting (if at all ) the image for enlargements ,noise wise?
And would you apply this before or after the enlargement was made.

Noise shouldn't be impact by the separation technique itself, though certainly sharpening based on enhancing the HF data will equally impact the noise (itself being high frequency).  If enlarging for print (I gather that's what you mean?) I would use this technique to sharpen after doing so, assuming that you're not meaning to output to a RIP, in which case I wouldn't resize at all.

Let me know if that makes sense; lots of pots on the stove right now.

Sean Baker wrote:
Noise shouldn't be impact by the separation technique itself, though certainly sharpening based on enhancing the HF data will equally impact the noise (itself being high frequency).  If enlarging for print (I gather that's what you mean?) I would use this technique to sharpen after doing so, assuming that you're not meaning to output to a RIP, in which case I wouldn't resize at all.

Let me know if that makes sense; lots of pots on the stove right now.

It sorta does, I thought so  too , as well on the applying it AFTER the enlargement is done. less artifacts to corrupt once it is enlarged (that make sense?)
I do mean printing , Ive had very little issue with it but on images with a higher ISO where the noise is a bit more 'annoying' (to put it lightly) .
Also is there a huge difference on 8bit -16 bit image enlargements .
Again I apologise if the question has been handled already or n/a.

ESCALANTE wrote:
It sorta does, I thought so  too , as well on the applying it AFTER the enlargement is done. less artifacts to corrupt once it is enlarged (that make sense?)

It does make sense, but the sharpening will function differently in such case, hence recommendation to do it at the final size.

ESCALANTE wrote:
I do mean printing , Ive had very little issue with it but on images with a higher ISO where the noise is a bit more 'annoying' (to put it lightly) .

If you're having trouble with the noise interacting with the sharpening algorithms, you can always try working on a separate, noise-reduced copy in order to generate your sharpening layer, which should afford you the best of both worlds.

ESCALANTE wrote:
Also is there a huge difference on 8bit -16 bit image enlargements .

Is there a technical one?  Yes, absolutely.  Is it one you're likely to notice from any but the best & best calibrated printers?  No.

Sean Baker wrote:
On real-time skin work:

Spawning off of what Panzerwolf started above, I've gone back to look more closely at what's going on with the High Pass filter proper and how we might exploit the new insights into its operation.  As stated, its method of applying the change can't handle a full-contrast scene as the range involved is sometimes too great for it to record.  I was hesitant to use PW's Curves techniques as they were imperfect for halving the contrast and recreating the scene perfectly (or at least to the standard previously set), despite countless attempts at changing curve values and playing with all manner of permutations, opacities, etc.  It should be noted that I don't believe this is the result of his efforts, but of limitations inherent to PS itself.  Curves didn't work; blending with 50% gray didn't work; blending with shifted grays didn't work (creating a 127.5 gray); but the Legacy Brightness / Contrast adjustment works. 

I know, I'm shaking my head over that one too, but let's use it:

1.) Take your image on which you want to do some skin work, creating a copy on which you'd like to work.
2.) Use Image->Adjustments->Brightness / Contrast; check the Legacy box; enter -50 in the Contrast box; 0 Brightness.
3.) Invert the layer.
4.) Convert the layer to a Smart Object.
5.) Run High Pass first, selecting your upper pixel limit for elimination.
6.) Run GB next, selecting the lower limit of elimination / upper limit of preservation in the detail.
7.) Tune to taste, rasterizing if you so choose for speed / size, mask, etc.

That Brightness / Contrast gives us a better result than any Curve or gray blend raises a lot of questions, but those will have to wait for another day.  In the mean time, you can do realtime skin bandstopping without loss of pixel-level quality (yes Virginia, you do retain 1/32k accuracy) .

Note: If you're not sure wth I'm talking about, please read  this within the thread where the mechanics for how these techniques apply to skin and are first demonstrated then explained.

*SIGH* Does this mean I need to write a new script and/or action set?

I'm always up for it, though. Just probably not today. I'll have to play with this method a bit first to see how it acts first-hand.

Sean Baker wrote:
On real-time skin work:

Spawning off of what Panzerwolf started above, I've gone back to look more closely at what's going on with the High Pass filter proper and how we might exploit the new insights into its operation.  As stated, its method of applying the change can't handle a full-contrast scene as the range involved is sometimes too great for it to record.  I was hesitant to use PW's Curves techniques as they were imperfect for halving the contrast and recreating the scene perfectly (or at least to the standard previously set), despite countless attempts at changing curve values and playing with all manner of permutations, opacities, etc.  It should be noted that I don't believe this is the result of his efforts, but of limitations inherent to PS itself.  Curves didn't work; blending with 50% gray didn't work; blending with shifted grays didn't work (creating a 127.5 gray); but the Legacy Brightness / Contrast adjustment works. 

I know, I'm shaking my head over that one too, but let's use it:

1.) Take your image on which you want to do some skin work, creating a copy on which you'd like to work.
2.) Use Image->Adjustments->Brightness / Contrast; check the Legacy box; enter -50 in the Contrast box; 0 Brightness.
3.) Invert the layer.
4.) Convert the layer to a Smart Object.
5.) Run High Pass first, selecting your upper pixel limit for elimination.
6.) Run GB next, selecting the lower limit of elimination / upper limit of preservation in the detail.
7.) Tune to taste, rasterizing if you so choose for speed / size, mask, etc.

That Brightness / Contrast gives us a better result than any Curve or gray blend raises a lot of questions, but those will have to wait for another day.  In the mean time, you can do realtime skin bandstopping without loss of pixel-level quality (yes Virginia, you do retain 1/32k accuracy) .

Note: If you're not sure wth I'm talking about, please read  this within the thread where the mechanics for how these techniques apply to skin and are first demonstrated then explained.

This uses Linear Light blend mode on that new Smart Object layer, correct?

Photons 2 Pixels Images wrote:

This uses Linear Light blend mode on that new Smart Object layer, correct?

Good point.  Yes.

Photons 2 Pixels Images wrote:

*SIGH* Does this mean I need to write a new script and/or action set?

I'm always up for it, though. Just probably not today. I'll have to play with this method a bit first to see how it acts first-hand.

OK. I lied. I got a quick action set together for this.

http://www.nunuvyer.biz/Photoshop/Actio … ration.atn

I hope I got this right.

Photons 2 Pixels Images wrote:

OK. I lied. I got a quick action set together for this.

http://www.nunuvyer.biz/Photoshop/Actio … ration.atn

I hope I got this right.

Looks perfect to me - save that there's no mask, my paintbrush isn't selected, I don't have white chosen, etc. [teasing]

Sean Baker wrote:

Looks perfect to me - save that there's no mask, my paintbrush isn't selected, I don't have white chosen, etc. [teasing]

OK, OK...updated. I shoulda known better anyway, huh?

PANZERWOLF wrote:
and in 8bit, the HP still creates moiré patterns, even with B/C

What type image are you seeing moire effects in?  I'd like to see if we can't work that one out, as obviously the 1/2 contrast thing requires old tools too .

Sean Baker wrote:
What type image are you seeing moire effects in?  I'd like to see if we can't work that one out, as obviously the 1/2 contrast thing requires old tools too .

your soccer guy image, radius 25
in 8bit it's actually worse with the contrast/brightness method
although it's not a real problem, since it's pretty low, with a maximum difference of about (1/2/1), but when amplified you can see moiré patterns that i haven't seen with the apply image methods

here's a comparison
left side is normal difference, right side is amplified with the steepest possible curve

PANZERWOLF wrote:

your soccer guy image, radius 25
in 8bit it's actually worse with the contrast/brightness method
although it's not a real problem, since it's pretty low, with a maximum difference of about (1/2/1), but when amplified you can see moiré patterns that i haven't seen with the apply image methods

here's a comparison
left side is normal difference, right side is amplified with the steepest possible curve

Is it Moire', or is it edge detail of a posterized image?