this frequency separation you all speak of...

[P_fuzz]

show me the way to it

j/k, I was wondering if there is an ultimate tutorial on this tech, so far I only found some really poor elaborated ones on youtube.

thanks

[P_fuzz]

and btw, how does it set itself appart (as a technique) from, say, DnB?

[Chain]

Frequency separation is a technique that accomplishes something different than d&b.

I recently covered it in this post; http://www.retouchpro.com/forums/pho...tml#post292871

Here are my actions so you can try and see how it works:
http://www.retouchpro.com/forums/att...1&d=1307869520

-- (copy from thread) --

Frequency separation basically turns your image into two layers. At the bottom is a layer containing the low frequency details (basically a blurred version of the image), and above that is a layer that contains the high frequency details (the "difference" between the blurred layer and the original). Blended together they should be identical to the original (slight differences might occur due to rounding errors and such).

This allows you to edit the two components separately. You can clone/heal fine details like pores without affecting the general shades in an area, or the other way around.

You can also decide to blur away medium size details afterwards on the underlying (low frequency) layer (probably best to do this on a duplicate layer not on full opacity, and with a mask). This can be called "Band Stop" as it removes a band of frequencies. Can be a great help when fighting "blotchy" skin.

You do not have to use Gaussian Blur (although this is the most common). You could use Surface Blur, Dust & Scratches, Lens Blur, or other filters as a base for the separation.

[Der_W]

The ultimate post about FS: https://www.modelmayhem.com/po.php?t...=439098&page=1

As to your second question: It's a whole different thing. D&B is used to selectively lighten or darken pixels. FS is used to separate your image into different levels of detail on which you can work individually without having to worry about the other levels.

Edit: The other Jonas was faster :-)!

[Chain]

Step by step if you want to try it manually (it is a bit cumbersome and I usually forget the Apply Image settings, that's why I use actions for it):

(assuming flat image)
1. Duplicate background twice. Call the bottom one LF, and top HF. (Hide Background as a backup).
2. Blur the LF layer until the finest details (those you want to separate out) just disappears. Maybe something like 2-10 px depending on the image.
3. Select the HF layer, and choose Image > Apply Image
4. Choose these settings for 8-bit images:

Layer: LF
Blending: Subtract.
Scale: 2
Offset: 128
Preserve Transparency, Mask, and Invert should not be checked.

5. Set the blending mode of the HF layer to Linear Light (100%).

The result (of the LF + HF layers) now looks identical to the original image. You can work on them separately (e.g. try blurring or airbrushing away unwanted shades on the LF layer). Works nice for straightening out fabric as well, not only skin tones.

For 16 bit images, apply image should have these settings:

Layer: LF
Blending: Add
Invert: Checked/yes
Scale: 2
Offset: 0
Preserve Transparency and Mask should not be checked.

[P_fuzz]

Yay, that was some quick response.

@Chain, I will try your action out, thank you.

[Chain]

PS: the actions work fine on layered documents.

[Der_W]

Originally Posted by Chain

For 16 bit images, apply image should have these settings:

Layer: LF
Blending: Add
Invert: Checked/yes
Scale: 2
Offset: 9
Preserve Transparency and Mask should not be checked.

Offset should be 0, not 9 ;-).

[Godmother]

maybe this helps http://nataliataffarel.tumblr.com/po...ng-and-cloning

[lunaxxxxxx]

but whats it all about that FS...?

[shift studio]

Originally Posted by lunaxxxxxx

but whats it all about that FS...?

Are you joking?

Chain explained what it is, and what its used for in his first post.
Then proceeded to explain how to do it, and link to a PS action to perform the FS
And it was explained further in the thread by very knowledgeable posters.

Yes, you must be joking - sorry.

--Shift Studio.

[Chain]

Originally Posted by Der_W

Offset should be 0, not 9 ;-).

*typo fixed

[P_fuzz]

had my first go, pretty cool, but still feels like i am faster at DnB than this with all the changing layers every now and then... guess i need to find my natural flow.

wonder if there was ever a retouchPro Live broadcast with someone using this technique from start to finish. would be an interesting one!

[Boneappetit]

Originally Posted by P_fuzz

had my first go, pretty cool, but still feels like i am faster at DnB than this with all the changing layers every now and then... guess i need to find my natural flow.

wonder if there was ever a retouchPro Live broadcast with someone using this technique from start to finish. would be an interesting one!

Hi:

Although the FS technique, achieves something different then D&B, as some other members stated above, most retouchers use it along with D&B. This technique is also used for some other things, like sharpening, blur, replace texture, hair retouch etc.

Here is a link to a simple video about it... http://www.youtube.com/watch?v=dMyae...79CF226CFE4BD0

[edgework]

Nothing's going to substitute for meticulous D n B, pixel-level cloning, targeted curves and judicious use of the healing brush. I've personally found the greatest benefit with split frequency layers is in smoothing out tones to a greater extent than is possible with the other tools. Like anything cool, it's easily abused, which is my gripe with the degrunge technique: it's almost impossible to use it without it calling attention to itself. With split frequency layers, you can build up the smoothing effect gradually, in real time. Like any good tool, it simply gives you more options.

[Siciliana]

I love frequency separation for work on skin that is really tricky. I especially love splitting it into 3--low, high light and high dark--this is a great way to remove little veins without removing the pores altogether. Of course you do really have to figure out when and where to put it in the workflow. I find it really useful for some things. Just another tool in the box, but I was delighted when I first discovered it. The Model Mayhem link Jonas provided is an excellent read on the subject.

[Der_W]

High light is the high pass with just the light information on it, high dark is just the dark information?
Seems like an interesting idea to play with, although I'm not sure whether it's really helpful (in my workflow) or not.

E.g. Why not just use "Darken" and "Lighten" modes?

[RobertAsh]

Originally Posted by Der_W

The ultimate post about FS: https://www.modelmayhem.com/po.php?t...=439098&page=1

Edit: The other Jonas was faster :-)!

Great article, very detailed, thanks for the pointer....

Originally Posted by Boneappetit

Hi:

Here is a link to a simple video about it... http://www.youtube.com/watch?v=dMyae...79CF226CFE4BD0

This is an excellent, easy to understand and follow video tutorial. It's where I learned about the technique. The video is straightforward and demystifying.

Also, don't get confused by the name -- the name Frequency Separation is completely inaccurate (although admittedly that name is very cool and sounds very sophisticated ). You are not "separating frequencies", you are separating texture from color into separate layers order to work on texture and color individually.

So after the tutorial and video, for me it's better to name my layers Texture and Color, which actually describes what I'm doing, not High Frequency and Low Frequency which don't really mean anything because Texture and Color aren't 'frequencies' of one another.

[Der_W]

Not true.
Frequency separation is a separation in terms of spatial frequencies, not a separation based on color and texture (texture is a word with different meanings anyways: https://www.facebook.com/note.php?no...29465173761662).

See these three posts from NIRBTG here: https://www.modelmayhem.com/po.php?t...1#post14971490

Color and texture are a completely different issue (and you'll see that in most cases the HIGH layer contains color as well if you bump up the saturation extremely).

[RobertAsh]

Jonas,

Cool. Thanks. Very useful information.

In practice, at least in the YouTube video, seems you're retouching textures on the High Frequency layer and blurring colors on the Low Frequency layer, is that always the case? If so, then I'll name my layers High Frequencies (Textures) and Low Frequencies (Colors). Otherwise I'll just learn more about how it's used and come up with some more meaningful names that fit better.

[Der_W]

This is one way to work, but some people also like to do a lower blur to create the split, so blurring the lowpass layer even more would look completely unnatural in these cases.

Of course you can name your layers in whichever way you want, just make sure to remember that the lowpass doesn't contain all the colors (unless you're doing an asymmetric split), because it could lead to problems concerning color spill if you work on the high pass like it didn't contain any color :-).

[Siciliana]

As for the 3 way frequency split, I learned it from the Model Mayhem post. (attaching screen capture) I had been using the other 2 way split before but I was concerned that it might be altering the final data in some way. When he said that the 3 way split was mathematically perfect, I decided to try it. I mostly use frequency separation to adjust value and tone blending on the blurred layer, but I found that, depending on the image, some blemishes are only on the high frequency dark layer, but there is still enough information on the high frequency light layer so you can remove the blemish on the dark layer and still have a very realistic look. For example, if a model had blackheads that were on the high freq. dark layer, you could lighten those up easily, by masking, but you would still have the highlights of the pores untouched in the high freq. light layer. It is worth playing around with to see what you can use it for. The other way is simpler, especially if you are planning to do a lot of healing of blemishes on the high frequency layer.

[Siciliana]

Oops, forgot to attach this.

[Der_W]

Ah, now I see :-)!
Of course the add/subtract method works perfectly and delivers the highlights and the shadows on separate layers, so it's better than my initial idea of using "Darken" and "Lighten" modes in terms of accuracy.

However is this workflow really comfortable for you or are you just using it on some images?
Personally I've never seen the benefits of working on a 100% accurate split if I on the other hand always have to switch layers and do healing on both high frequency layers.

Edit: If it's just about the accuracy, you can improve it with this method while retaining the benefits of a single high pass layer.
Just put a 50% gray layer underneath your high layers, set their fill opacity to 50% and merge with the gray layer.
If you're working in 16bpc, you'll still need to place a curves adj. layer with 60% opacity on top clipped to your high layer with the points (output/input): (0/2), (253/255).
Little more accurate in both 8 and 16bpc but still just one layer :-).

[Boneappetit]

Originally Posted by RobertAsh

Jonas,

Cool. Thanks. Very useful information.

In practice, at least in the YouTube video, seems you're retouching textures on the High Frequency layer and blurring colors on the Low Frequency layer, is that always the case? If so, then I'll name my layers High Frequencies (Textures) and Low Frequencies (Colors). Otherwise I'll just learn more about how it's used and come up with some more meaningful names that fit better.

Hi: IMHO, technically speaking Jonas is right, but I can understand why you understand it using the other terminology, and that would be because "frequency" is an "audio term", not a painting or drawing term as far as I know, (I could be wrong thou...) Like, Bass notes (low frequencies) and a flute notes (high frequencies) also like a woofer and a tweeter... and there are also the mid range tones like a tenor sax or a trombone...

[Siciliana]

Jonas, I have not made the 3-way split a regular practice, but I've played with it a bit and think I could think of more ways in which to make use of it when dealing with certain problems over large areas. As a painter in traditional media, I would eventually like to translate my photoshop skills back into doing digital artwork. I can think of ways in which this split could be very useful to me in achieving effects. I am new to retouching anyhow, so I haven't really developed my workflow habits in stone... still figuring out what works best for me.

[Chain]

Originally Posted by Boneappetit

...because "frequency" is an "audio term", not a painting or drawing term as far as I know, (I could be wrong thou...)

Yeah, the word "frequency" is more often used in relation to things like audio, and not really encountered in painting and drawing. With digital images I've only encountered it in the case of frequency separations and FFT.

It is the correct term, although I can understand if some people like to think of it as texture/tone instead of high/low frequency.

[RobertAsh]

Originally Posted by Boneappetit

Hi: IMHO, technically speaking Jonas is right, but I can understand why you understand it using the other terminology, and that would be because "frequency" is an "audio term", not a painting or drawing term as far as I know, (I could be wrong thou...) Like, Bass notes (low frequencies) and a flute notes (high frequencies) also like a woofer and a tweeter... and there are also the mid range tones like a tenor sax or a trombone...

Originally Posted by Chain

Yeah, the word "frequency" is more often used in relation to things like audio, and not really encountered in painting and drawing. With digital images I've only encountered it in the case of frequency separations and FFT.

It is the correct term, although I can understand if some people like to think of it as texture/tone instead of high/low frequency.

Exactly right. But Jonas' url has a post that explains it's not color frequency or anything like audio frequency the way I was thinking - it's spatial frequency.

That makes perfect sense. Because you're separating your laters into:

• Attributes that change very frequently as you pass over the surface of the image -- like tonal changes that define skin pores or hair, etc. (i.e., the frequency of change is often or high, aka high frequency items or attributes), which are typically related to texture.
• Attributes that change less frequently -- like changes in color, highlights and shadows over broader areas like an entire face (i.e., the frequency of change is seldom or low, aka low frequency).

[Chain]

(Nobody said we separated "color frequency", although I guess we could if we wanted to).

It is indeed spatial frequency as the last post clarifies. Nice post for those who were still confused.

[John Wheeler]

Originally Posted by Boneappetit

... that would be because "frequency" is an "audio term", not a painting or drawing term as far as I know, (I could be wrong thou...)

Not used as commonly in the graphics arts field yet used often by Mathematicians, Physicists, and Engineers (may give a clue about the background of the person who came up with the technique and coined it).

Anything that changes with regular periodicity per any parameter can be characterize by frequency. With time it is called temporal frequency, with distance it is spacial frequency and can also be used in combination of time and distance (referred to in cases such as removing falling snow in a video which changes in both time and position).

Link on Wikipedia for spacial frequency: http://en.wikipedia.org/wiki/Spatial_frequency

Maybe a more interesting link (well for some) is how the eye's ability to recognize items of different contrast is also a strong function of spatial frequency of what you are observing. The eye's sensitivity drops off at both low and high spatial frequency (not unlike the ear with audio/temporal frequency): http://vision.psy.mq.edu.au/~peterw/csf.html

And right now you may be wondering how this thread got changed to the Science Channel