Extended frequency separation

[fraiseap]

I have been playing around with frequency separation using a couple of methods;

1. Using surface blur to create a low frequency layer and using high pass to create a high frequncy layer and blending the two with linear light

2. Using the Model Mayhem technique of creating a low frequency layer using Gaussian blur and using "apply image" to subtract this from the original to create a high frequency layer.

My concern is that 1. uses high pass which is not exactly the mathematical equivalent of <original minus low frequency> so I am losing data. In contrast 2 does not get rid of the grunge.

So that got me wondering if I should modify the Model mayhem technique to create a low, middle and high frequency layer which when combined is an accurate representation of the original. I can then tidy up the middle frequency layer using clone tool, healing brush and D&B. I can also reduce the opacity of the middle layer if required.

Has anyone tried this? Is it the same as the quick degrunge or not? Is it worth the effort?


I should add that I know D&B is superior but, as a working portrait photographer I cannot spend hours on an image so I need something quicker but effective.

[Der_W]

Re: Extended frequency separation

You can use Surface Blur for the low pass and the apply image to create the according high pass, that won't change anything with the accuracy :-).

[Flashtones]

Re: Extended frequency separation

You can use Apply Image on the Surface Blur same as if you used GB

If you want to split the image on multiple frequencies Sean has made a PS script that allows you to split the HF data across as many layers as you want. For instance, if your Blur radius is 10 px, you can ask for that to be split among as many HF layers as you like, so if you choose 5 layers each would span 2 px of HF data.

That's probably not clear, but check the links on the opening post of the HPS thread and you should see links to the action and Sean's description.

[fraiseap]

Re: Extended frequency separation

Thanks for that. It is good to know that surface blur can be used as I prefer that (it avoids bleed from lips, dark hair etc).

I think the actions to split the high frequency layer is exactly what I need. Where do I find the post you referred to?

[Flashtones]

Re: Extended frequency separation

I think the actions to split the high frequency layer is exactly what I need. Where do I find the post you referred to?

The model mayhem high pass sucks thread, which I don't currently have a link for. Just google it. the opening post contains many links worth exploring.

[fraiseap]

Re: Extended frequency separation

Thanks,

for those reading this, here is the link

http://www.modelmayhem.com/po.php?thread_id=439098

[Flashtones]

Re: Extended frequency separation

Scanning that first post briefly, I don't see a link to Sean's script. Hopefully someone can direct you, but really, you should take the time and read the whole thing through. It's time well spent.

[fraiseap]

Re: Extended frequency separation

OK, I have now read most of the MM thread and it is very informative (especially if you ignore the "you are destroying art" type rants). I have also found Sean's blog which is extremely good

http://sean-blog.twicebakedphoto.com/2010/08/visual-frequencies.html

Now I have a thought. You could create 3 layers - a low frequency layer by applying blur (Guassian, median or surface) a high frequency layer (either by using high pass or subtracting the blur layer from the original as described by Sean) and a middle layer by applying a bandpass filter between the frequencies of the other two layers. This would allow me to do what I wanted (edit the middle -or grunge- layer without affecting texture or shape)

I tried this and the combined image looked OK but had too much contrast. So my question is, should I reduce contrast of the top two layers to 50% and use linear light for both? Or have I missed something?

[Cron]

Re: Extended frequency separation

You need to make the layers in order low -> middle -> high. The second blur should be made to the (to be-)middle frequency layer.

You'll need a "help-layer" of Low+Mid after the second blur to make the high freq layer.

1. Copy background 2 times, name them from bottom to top: Low, Mid, High
2. Blur Low and apply Mid normally and set to Linear Light
3. Blur Mid and Copy Merged (Low+Mid) the resulting image to a layer and name it Help
4. Apply High using Help (instead of Mid) and set to Linear Light
5. Hide/trash Help

[fraiseap]

Re: Extended frequency separation

Brilliant - thanks. All working now.

[Der_W]

Re: Extended frequency separation

Too much trouble imho.

Use two duplicates, LOW and HIGH, blur the LOW, use apply image on the HIGH.
Now copy the LOW layer, name it MIDDLE, blur the LOW and use apply image on the MIDDLE.
You can do this as often as you wish and there's no need for Help layers ;-).

[fraiseap]

Re: Extended frequency separation

Too much trouble imho.

Use two duplicates, LOW and HIGH, blur the LOW, use apply image on the HIGH.
Now copy the LOW layer, name it MIDDLE, blur the LOW and use apply image on the MIDDLE.
You can do this as often as you wish and there's no need for Help layers ;-).

Thanks, I will try that. I assume you use linear light blend for the middle and high layers.

[fraiseap]

Re: Extended frequency separation

Too much trouble imho.

Use two duplicates, LOW and HIGH, blur the LOW, use apply image on the HIGH.
Now copy the LOW layer, name it MIDDLE, blur the LOW and use apply image on the MIDDLE.
You can do this as often as you wish and there's no need for Help layers ;-).

I tried this and you can remove the lower frequencies from the MIDDLE layer but the high frequencies remain . As they are also on the HIGH layer you have 2 copies of the high frequencies resulting in an oversharpened look. The solution is to apply gaussian blur to the MIDDLE layer after the "apply image step"

[Flashtones]

Re: Extended frequency separation

I tried this and you can remove the lower frequencies from the MIDDLE layer but the high frequencies remain . As they are also on the HIGH layer you have 2 copies of the high frequencies resulting in an oversharpened look. The solution is to apply gaussian blur to the MIDDLE layer after the "apply image step"

For my own education, can you help me better understand what you are trying to achieve that, say, these more common methods don't allow?

http://www.modelmayhem.com/po.php?thread_id=439098&page=5#post9709595

http://www.model-citizens.com/Tutorials/Separation-1.wmv

Also, did you get that script from Sean? In the meanwhile take a look at this:

http://retouchpro.com/tutorials/?m=show&id=149

[fraiseap]

Re: Extended frequency separation

They are some great links - thanks. Especially the first one.

I am not sure I am a position to educate anyone but I will try to explain what was going through my head.

I spent several months trying to smooth skin using blur and reducing opacity but hated the plastic look I got. I then tried reintroducing noise to give a bit of texture but that still looked awful. I then discovered applying a High Pass layer with linear light over a blurred layer, masked and with opacity reduced. This has served me quite well for a few months.

Recently I discovered the High Pass Sucks thread and have tried to understand frequency separation. I love the fact that the combined layers result in the original image but was struggling with the fact that the stuff I wanted to get rid of (spots, pimples, skin grunge) was on both layers. I wanted to create a low frequency layer which was very smooth without blemishes and a high frequency layer with perfect pores (I dont mind healing a few zits etc but i wanted that layer to be pretty much good to go). That would (in my mind) leave all the other data in a middle layer which i could clone, heal, reduce opacity to get the effect I wanted.

The problem was that all the tutorial and actions I have found only create 2 layers - high and low frequency - hence my post.

I hope I have explained that OK

Adam