Great retouch video using Frequency Separation

[julianmarsalis]

From ModelsMayhem:

Originally Posted by FLEXmanta

Freq split tutorial by me on hiendworkshops. Sorry for compression, i delivered HD. Also, it's hard tu put it all in a <5 min video.

High Pass cloning tutorial on hiendworkshops

Hope it helps.

Originally Posted by FLEXmanta

Yeah, I recently got offered to teach at a photo academy here in Madrid starting September. It's gonna be my first experience teaching post and digital teching.

I also have another 20 minute HD video on frequency separation, but it's in Spanish. Still, i think you'll get the picture by just watching.

http://www.ojodigital.com/foro/retoq...ecuencias.html

Originally Posted by FLEXmanta

Ok, here's a small example.

3 Layers:
-Low
-Painted corrections in normal mode
-High


http://www.omarjosef.com/clients/discol.psd

The unsplit image was intentionally desaturated under the eye just to make a point.

Really brings home how powerful a tool frequency separation is for retouching/fixing images.

[pelez]

nice share.....thank you

[mistermonday]

Julian, the last link does not work. It looks like an HEX-ASCII representaion of the file.
Regards, Murray

[julianmarsalis]

I had a similar issue with the file from flex and used save as and made sure it named the file with the .psd extension and it worked for me.

[mistermonday]

Julian thanks, that worked.
Regards, Murray

[amica999]

Thanks for sharing!

[kav]

Regarding the video, I don't find it practical to blur out things on a "fashion photo" as he suggested works for a nice effect. The other thing being, his cloning was really sloppy. He shows his cloning layer at the end and it has huge broad strokes all over it. When you're not experienced with working on skin sometimes this might seem like the way to do it, because it gives you a lot of feathering and seems to maintain the lighting at first glance, but it will push the texture around a ton, even done in his detail separated method.

For minimal impact (and it's easy) i'd suggest either a background copy or blank layer set to lighten or darken. With blank layers you end up having to do all of your lightening or darkening on one, then if you want the other blending type it must be on a higher blank layer. The only real advantage is that it allows you to see where your brush went down. It doesn't blur any less or any more than the background copy, and both are easily editable given the use of history brush or eraser + re-duplication of the background.

[julianmarsalis]

I think your missing the basic point of the separation is doing for you or maybe you haven't done this yourself if you think this

but it will push the texture around a ton

you don't touch the texture on the texture layer that way.

The key is you can actually use cloning or flat out painting on the layer between the HF and LF layer without touching the texture at all. Heres another good video I think your missing the point entirely with the statement your making:

http://www.model-citizens.com/Tutori...paration-1.flv

Originally Posted by Mask Photo

tl:dnr (but the number of trolled out posts was interesting)

my method: paint the right color into the low frequency layer (or blur the blotchiness away, or a combination of the two)

while it's a lot of work to split the frequencies, i do it with every photo i process by default, so as long as i'm there, it's easy to fix the skin...

http://www.modelmayhem.com/po.php?thread_id=576703

Originally Posted by Julian Marsalis

The first link has a added comma in the link http://www.model-citizens.com/Tutori...paration-1.flv

http://www.model-citizens.com/Tutori...paration-1.wmv

I am a huge fan man retouching and restoring images are made so much simpler once you understand simply where and how to retouch the different layers. I started using a surface blur on the color and low pass layer and love the results.


Here's a similar tutorial from retouchpro:

[julianmarsalis]

The high Frequency layer contains the texture. The low frequency layer contains the color and tone. The middle retouch layer allows correction of tone and color without touching the texture use any tools that can be set to current layer and below. The texture can be fixed with using and tools set to current layer. It's a great fast way to edit around the problems with texture and color cloning artifacts. Has many uses in my book and is now a stable of my work flow editing images has never been this fast and easy before....

[kav]

I'm going to try that a bit more seriously. The issues it solves though aren't really my main problem.

[flexmanta]

Originally Posted by kav

Regarding the video, I don't find it practical to blur out things on a "fashion photo" as he suggested works for a nice effect. The other thing being, his cloning was really sloppy. He shows his cloning layer at the end and it has huge broad strokes all over it. When you're not experienced with working on skin sometimes this might seem like the way to do it, because it gives you a lot of feathering and seems to maintain the lighting at first glance, but it will push the texture around a ton, even done in his detail separated method.

For minimal impact (and it's easy) i'd suggest either a background copy or blank layer set to lighten or darken. With blank layers you end up having to do all of your lightening or darkening on one, then if you want the other blending type it must be on a higher blank layer. The only real advantage is that it allows you to see where your brush went down. It doesn't blur any less or any more than the background copy, and both are easily editable given the use of history brush or eraser + re-duplication of the background.

Seems like you didn't understand the purpose of the frequency separation.

[Flashtones]

I still think some good tutorials on selecting the proper separation radius are in order. I'm constantly experimenting and still don't feel I have a handle on it. Too high a radius puts too much tonality into the HF layer, rendering the low pass layer less flexible. Too low a radius leaves texture feeling "delicate" and not robust enough; subject to getting lost.

Flex and P2P have both done a nice job of showing how and why to separate, but I think the topic is worthy of some more extensive demonstrations.

1. When in the workflow to separate (ie, before or after basic cleanup).
2. Proper radius to separate at, and an exploration of differing radii for different textures or intensions.
3. Things one can do with a separated image. For instance, for degrunge I prefer to blur the lowpass layer (or portions of it) over inverted highpass, because I get less edge blooming than IHP. And other techniques, like clipping a curve to the HP layer. Etc.
4. Uses on items other than skin.

The why, how, when (or when not), where, and to what effect of it all.

[julianmarsalis]

I point you to the tread of thread that started this all one heck of a read High Pass sucks lol http://www.modelmayhem.com/po.php?thread_id=439098

Originally Posted by SRB Photo

Summary Information and Links now at bottom of post

Bottom Line Up Front: High Pass is an inaccurate spatial-frequency gaussian based separation technique. The steps below and the actions provided effect a much more accurate result than can be accomplished otherwise.

Background: Ultimately, this is a spinoff of Mr. Connery's thread on deconvolution as an image sharpening technique here, wherein I set forth my method for finer control of such while both remaining within Photoshop and not spending additional $$ on plugins. Key to its success is the separation of high and low spatial frequency image data and the fine-tuning of each.

Separation of spatial frequency data has a number of applications in image editing. Whether for the oft-suggested "High Pass Sharpening", to recover detail lost in an OOF / moving image, or to enhance local contrast throughout the image, the accurate separation of frequency data is relevant for anyone who is a stickler for image quality.

Findings / Technique: In my own experimentation, I've found that HP gives exceptionally high differences from 'truth' when separating high and low frequency information. The proposed alternative technique for working with 8bit image data is as follows:

1.) Start with two copies of the image to be separated.
2.) Working on the bottom copy, run the gaussian blur filter at the intended pixel frequency (same as you would input into the HP filter).
3.) Selecting the top copy, choose Apply Image from the Image menu at top (Shift+Ctrl+A on PC; Cmd+Shift+A for Mac).
4.) In the Layer dialog, select the bottom layer which you blurred in step 2.
5.) In the Blending dialog, choose Subtract.
6.) Enter '2' into the Scale box, and '128' for the Offset.
7.) Preserve Transparency, Mask, and Invert should not be checked.
8.) Choose OK.
9.) Your top layer will now look much as a HP result, albeit a bit flatter. Set the Blend Mode to Linear Light. Opacity should remain at 100%.

Working in 8bit mode, this will give you an accurate frequency separation to about 129/32k or 1/256.

Working in 16bit mode, however, we encounter a problem with the offset factor applied in Step 6 above (50% grey in a 16bit system cannot be expressed by an 8bit number). This can be overcome using the alternate technique which follows:

1.) Start with two copies of the image to be separated.
2.) Working on the bottom copy, run the gaussian blur filter at the intended pixel frequency (same as you would input into the HP filter).
3.) Selecting the top copy, choose Apply Image from the Image menu at top (Shift+Ctrl+A on PC; Cmd+Shift+A for Mac).
4.) In the Layer dialog, select the bottom layer which you blurred in step 2.
5.) In the Blending dialog, choose Add.
6.) Check the Invert box.
6.) Enter '2' into the Scale box, and '0' for the Offset.
7.) Preserve Transparency and Mask should not be checked.
8.) Choose OK.
9.) Your top layer will now look much as a HP result, albeit a bit flatter. Set the Blend Mode to Linear Light. Opacity should remain at 100%.

Conclusion: Accuracy for this technique is fantastic, with a maximum difference from the original of 1/32767. Unfortunately, this technique is not valid while working in 8bit mode as it doesn't have a perfect middle gray as 16bit does.

Afterword: Walking through all these steps can get a bit tiring, and I've created an action which I'll link below to run you through them. It works in a separate window utilizing the 16bit technique above to retain an accurate separation, but, as I'm not an Actions Wizard, it will not move the data back into your original document for you. If someone more adept than I would like to add that in I will gladly update this post; otherwise it remains to the user to bring both or either layer into your document for use.

What will you do with it?
- "High Pass Sharpen"?
- Create a high-spatial-frequency mask?
- Create a low-spatial-frequency mask?
- Something else? Post your results and findings here. It's time to up the ante on sharing what we find.

Action

Please Note: As outlined above, this technique is designed to end with the image on screen looking exactly the same as it did when you started. The difference is that your image data is now separated onto two separate layers (separated by size of detail - spatial frequency), allowing you to edit them independently. If you want to apply "high pass sharpening" with this technique, simply disable the low frequency ('blur') layer and it will appear immediately.

------------------------------------------------------------------
| Links to subsequent valuable information & downloads |
| contained within the remainder of this thread. |
------------------------------------------------------------------

JeF Briguet describes a similar procedure to reduce moire here.
grahamsz discusses healing at different spatial frequencies for skin evening here.
grahamsz provides a visual explanation of spatial frequency here.
Photons 2 Pixels' first skin smoothing action can be downloaded here; instructions here.
grahamsz elucidates one of the best and natural skin smoothing techniques around here.
Photons 2 Pixels' second skin action set is here and mentioned here.
syd47421 posted his implementation of a PS 'equalizer' here, discussed here.
Photons 2 Pixels' multistage separation actionset can be obtained here, discussed here.
I post a simple noise-generation script here and discuss its purpose here.
Photons 2 Pixels' third generation skin & sharpening actions are here, discussed here.
Photons 2 Pixels makes my noise script user friendly and replicable here, discussed here.
Koray demonstrates a series of derivative local contrast enhancements here and explains it here.
Panzerwolf laid the foundation for describing what's wrong with the HP filter, which I explain here.

For authors: If your contribution is not listed here and should be, or if I've listed something incorrectly, please send me a note so that I can keep this up to date and accurate. Thanks!

[KTG]

Flex,
I watched the Spanish language version and could follow it for the most part. But it would sure be nice to have an English version of that video. The short English version is excellent though.

k

[kav]

Originally Posted by julianmarsalis

I point you to the tread of thread that started this all one heck of a read High Pass sucks lol http://www.modelmayhem.com/po.php?thread_id=439098

I've never liked the high pass filter much for pretty much anything. For me it was basically for solving weird problems. Say you dodged out a large wrinkle on a dress and that previously shadowed area lacked proper texture (sometimes it's just underexposed in the shadows and noisy), you could use the high pass to hijack some texture from another area. I'll probably switch to using this instead for the same effect as it comes out a bit better. Another use I could see would be doing something similar on skin areas that are too lacking in texture. Sometimes it's just the makeup and lighting and looks weird in a small area. There are a lot of limitations on how far i'd want to go on the lower layer, but this does seem to have some advantages for quick cloning/cleanup. I'd suggest setting a snapshot right after you make the 2 layers so you can easily history brush back if you notice a mistake. I wish it inflated the file size less Some of these files are 200 mb + in 16 bit before I start adding layers.

[julianmarsalis]

That's why the High Pass sucks was started and most now use apply image from a Guassian blur layer to create a High pass layer