Blur / High Pass Radius

I use blur and high-pass a lot in Photoshop,in fact they are the base of my most successful actins and filters.

In Photoshop an image can be separated into a "High Pass" part and a "Low Pass" part by applying blur and high-pass at the same radius.
(They can be combined later by using Linear Light @ 50%)

However, in FF none of this works. The radius (ok, percentage) of one filter seems to bear little relationship to the percentage input of the other.
To mix things up it seems that the high-pass contrast adjustment also comes into play.

For example, to make a complementary filter to a blur of 10 (percent), the closest I can get is a High-Pass with radius of 44 and a contrast of 46.

Is there a magic way of making these paired filters, or is it "FF just doesn't work like that" ?

(or am I dong something very silly)

Rô

High Pass and Blur.ffxml

Hmmm, actually you can do what I did with FF's highpass filter too.
That is, with HighPass's radius at 50 (and contrast 0) in the foreground of the linear light blend (with 100% opacity), and the Gaussian Blur at 10 in the background this seems to remake the original image.


highpassblur2.ffxml

Ken,

Quote
In Your examples Linear Light is set to 100%. It should be set to 50% (128) to add or subtract. (or maybe I am wrong?)

Yeah, I was fudging it.
(I thought it was interesting and hopefully maybe helpful in some way. ...I guess it at least helped me figure out FF’s Blur and High Pass radii work at a 1 to 5 ratio. I’ll attach the “proof” below.)

I guess we need to figure out what exactly PS’s High Pass filter does in order to hope recreate the event correctly…

highpassvsblur.ffxml

For kicks, I made a quick test-filter so people can test how similar/different their Photoshop/PSP blend modes work compared to FF's.


Here's the instructions:
1. Make one layer in your PS/PSP/etc. program have the first perlin noise image in the filter. (I.e. be sure to have the "Original, Your's, FF's, Difference" controller set to 1 and apply the result to your layer).
2. Back in PS/PSP/etc., duplicate that layer, and blend the two of them with your blend mode and opacity of choice. Then merge the two layers into one.
3. Take that layer back into the test-filter. Set the blend mode and opacity to what you used in step 2.
4. Set the "Org,ur's,FF's,Difrnc" contoller to 4 to see the Difference blend result. The more black the result is, the more similar your graphic software's blend is to FF's blend.
4b. The white point adjusts the sensitivity of the test, where the higher it's value is, the blacker the results should be.


I found FF's Linear Light blend at 50% opacity to be somewhat close to PS3beta's -- where the result looked pure black with the white point at 5, but with a lower white point the image becomes less black.






blendmodestest.ffxml

Hi Vladimir.

Ssamm's snippets showed the 5:1 ratio very well

I open a new image 1000x1000 pixels
Plug it into a Blur component set to 10% (this equates to 100)
Plug it into a High Pass set to 50% (this equates to 100)
And blend them back together with Linear Light at 50%

The results are not the same as in PS7.
Adding High Pass contrast (40) helps but Ro's figures are still closer.

I also tried the Linear Light formula from here
http://www.filterforge.com/forum/read...essage1762

And got exactly the same results,


Ken.

Hi Rô.

I thought I had RTFM

But I can't find this

Quote
For the linear light use 100% and not 50%.

It does make sense and solves the problem but I can't find it in the Manual?


Ken.

I'm glad I found this topic.

I created a workaround for what you do with FF. By the way, FF crashes my cs4 PS.
I couldn't attach it here because the many layers inflating the file (I never imagined using all copies of the same smart object no-composite-for-compatibility-saving would make the file so big!!. It seems like mart objects are not smart enough yet)
so here it is http://socialblogsitewebdesign.com/Hi...eation.psd

I think the problem is the high pass is calculated with a "quick" gaussian blur (like the ones that show squared patterns) and the gaussian blur you apply to the photo is "optimized", making some differences.

I replicated the highpass effect with filters (like 10 layers with different grouping and blending modes) and the result is EXACTLY the same than the input image (well, except in the tones close to completely white areas)

I noticed it works better with "median" than with "Gaussian Blur". It extracts more detail close to the middle tones than using blur. Again, I think the gaussian blur uses some kind of "soft" edge of action, or so.

Can you figure out a way to come up with the calculation used by the high pass and create a median based high-pass filter instead of gaussian blur based?

Thanks.