High Pass filtering 2.8 vs. 2.10

[Espermaschine]

The manual way of making a High Pass Filter (that is also used in RobA's script) is:

1. duplicate image
   
2. desaturate
   
3. invert
   
4. reduce opacity to 50%
   
5. gaussian blur
   
6. merge down
   

When you do the steps in 2.8, the image becomes 50% gray at step 4.

This is not true for 2.10, unless you set the Blendmodes to Legacy.

I know Gimp 2.10 has a GEGL High Pass Filter, but i want to understand the "wiring under the board".

The only 2.10 procedure i can up with is:

1. duplicate image
   
2. desaturate
   
3. invert
   
4. set layermode to 'Grain Merge'
   
5. gaussian blur
   
6. merge down
   

So what is going on ? What has changed that makes the old way dysfunctional ?

[Ofnuts]

1) Define "invert". Even in 2.8, you have "INert" and "Value invert"
2) Working on a gamma corrected image or on a linear light one?
3) I don't see much of a highpass filter in the technique shown (with just a "Normal" mode). Grain merge is better

On the whole, 2.8 works directly on the gamma-corrected values (with a few exceptions, and I expect Color->Invert to be one). In 2.10 it works mostly on the linear values and if you want something that looks like the gamma-corrected processing you use the legacy modes.

[Espermaschine]

I cant answer any of that.

For what i do with Gimp, having so many options for doing something simple as desaturating or inverting is not an improvement.
Maybe it has a lot of relevance for people who edit photos.

I like a High Pass filter because its non destructive. In 2.8 this requires to apply the blur factor by gut feeling.
Theoretically 2.10 gives me more control/convenience with its live Blur filter (when doing it the manual way).

I guess i will have to stick with the GEGL High Pass, then.

[rich2005]

I did not really recognise that RobA Gimp 2.8 procedure. Going back to meetthegimp episode 164 ( from 2011, I have them archived )

The procedure is duplicate -> invert -> grain merge That gets the grey layer & Gaussian blur brings out the detail. That is the same as your Gimp 2.10 process.

The only different version I know is here https://opensource.photography/content/c...p-and-gmic and that uses grain extract.

Duplicate -> Gaussian Blur -> Layer mode Grain Extract -> New from visible.

Neither of these use reduce opacity (50%) of the inverted layer.

[Espermaschine]

(10-15-2019, 05:09 PM)rich2005 Wrote: I did not really recognise that RobA Gimp 2.8 procedure.

He describes the procedure on his web-blog:
http://www.silent9.com/blog/archives/152...lugin.html

Quote:They are all straightforward, using the basic "duplicate, blur, invert, 50% opacity, merge down" process

You can easily test it yourself in 2.8:
duplicate image
desaturate
invert

the image does not turn gray unless you reduce the opacity to 50%. And the details will come out after you applied the Gaussian Blur.

I guess there are several ways to skin a cat. In the end what matters is to filter out all of the low frequencies so you are left with the details.

[rich2005]

Quote:You can easily test it yourself in 2.8:
duplicate image
desaturate
invert

the image does not turn gray unless you reduce the opacity to 50%. And the details will come out after you applied the Gaussian Blur.

I did indeed 'test it myself in 2.8' which prompted my comment, because duplicate layer, desaturate, invert gives this. https://i.imgur.com/CgP14nb.jpg Maybe some procedures missing.

Worth a look at the meetthegimp video. Rolf is a bit verbose but it is a good show. https://blog.meetthegimp.org/episode-164...harpening/

[Espermaschine]

(10-15-2019, 06:36 PM)rich2005 Wrote: Maybe some procedures missing.

see my first post

   

[rich2005]

(10-15-2019, 06:47 PM)Espermaschine Wrote: see my first post

I did, and that is not the same as you wrote.

duplicate image desaturate invert reduce opacity to 50%

is not desaturate duplicate invert....

[Ofnuts]

(10-15-2019, 04:38 PM)Espermaschine Wrote: I cant answer any of that.

For what i do with Gimp, having so many options for doing something simple as desaturating or inverting is not an improvement.
Maybe it has a lot of relevance for people who edit photos.

I like a High Pass filter because its non destructive. In 2.8 this requires to apply the blur factor by gut feeling.
Theoretically 2.10 gives me more control/convenience with its live Blur filter (when doing it the manual way).

I guess i will have to stick with the GEGL High Pass, then.

There is also the "freaky details" method.

[tmanni]

(10-15-2019, 04:13 PM)Ofnuts Wrote: On the whole, 2.8 works directly on the gamma-corrected values (with a few exceptions, and I expect Color->Invert to be one). In 2.10 it works mostly on the linear values and if you want something that looks like the gamma-corrected processing you use the legacy modes.

You don't need to use the legacy modes. Open the layer properties dialog, and change the Composite space from AUTO to RGB (perceptual).

(04-24-2020, 05:05 PM)AlliPlatt Wrote: Hi....one difference might be in the GEGL version used for the build. Could you tell me which BABL/GEGL versions are associated to Gimp 2.10.10 in Arch?
For the AppImage I am taking the GEGL git HEAD, which is probably not a good idea for release builds.

Here, nothing related to the BABL / GEGL releases used ; I always use the last developpment releases of BABL/GEGL compiled manually and it is really stable.