Odin,
did you try it embedded into a script?
I forgot to say that almost all my "help" requests come from trying to use a gimp filter from inside a script, not interactively; anyhow thanks, I will try again.

I wanted to see how Border Average filter behaves interactively.
I got some "surprising" results:
- create a new image
- filled with ONLY PURE GREEN
- launched Border Average with default values (3,16): result = color with 8red+248green+8blue (!?)
- changed the bucket size to 64: result = color with 32red+224green+32blue (!?)
- changed the bucket size to 256: result = color with 128red+128green+128blue (!?)

So it seems that the filter:
- takes half the bucket size as a "distributed" value all over R,G,B
- then takes the same value and "deducts" it from the real value found (the real color is 0 255 0)

To double check I changed the color of the image to PURE RED (255 0 0)
Results confirm my hypothesis.

Could someone check my results?
and, if confirmed:
does someone know what is wrong with the filter?

I'm running Win 10 with Gimp 2.8.16

Thanks

Why don't you check the source code? After all this is the great benefit of Open Source Software.

where do I find it? and -even more important!- is it written in C? in this case I cannot "inspect" nor amend the code

BTW I wrote my own filter, to use until the "official" is not checked/updated

I didn't try it within a script; I meant the option choice(s) can change the output exponentially.

Odin
did you now try to double-check what I noticed also in the INTERACTIVE use?
that "strange" calculation....

You can read the doc, it's fairly explicit about how it works and why results can be surprising.

The doc can't explain how a TOTALLY GREN or a TOTALLY RED layer returns that strange value.
It explain how the results can look strange having MORE than one colour in the layer.
But, in any case, I keep my own filter which works perfectly IMHO (returning the REAL average of the border size selected)

I used a 400x400 solid color 215eb8.
Every number of colors i selected resulted in the fg being changed to 215eb8. 3 px border btw.

thanks Rod!
Thus it's my PC.

edit: anyone else would like to contribute with this experiment?

dinasset

Firstly, you have three channels of color red, green, and blue that are individually treated as grayscale values (0 to 255). For each channel, all the values existent in the border region are collected in groups of the specified bucket size.

For example, if the bucket size is "8" then there will be 32 different bins which hold the pixel value counts in the ranges:
0-7 # First bin
8-15 # Second bin
16-23 # Third bin
:
:
240- 247 # 31st bin
248- 255 # 32nd bin

For your "pure green" color, all of the red values would appear in the first bin, since the red component of a pure green is "0". Likewise for the blue values. All of the green values would get collected in the 32nd bin, since they'd be "255".

Now for your example it's pretty obvious that the first bin has only counted the "0"s and thus it'd seem strange to use the midpoint value of the bin (i.e., "4") as the bin's associated value -- but what if the color was not pure and the first bin for the red component contained an assortment of values in the 0-7 range? Then it would make it entirely reasonable to use the midpoint.

Likewise for your case of the green channel. All values actually equal "255" and get counted into the 32nd bin, but once placed there the original values are "forgotten" (presumably to simplify calculations and reduce storage) and there's little recourse but to use the midpoint value of the 32nd bin (i.e., "252") as the computed channel value.

Hopefully, you see through my obfuscated explanation to the basic idea of the cause of your anomalous colors. Sorry I did not explain it well but I'm in a bit of a hurry.

Thanks, Saulgoode, you explained it perfectly!
Now it's clear.
So, I will never try to use it and will stay on my script, which calculates the exact mean values.

While discussing average colour sampling, I wrote this script a way back: http://registry.gimp.org/node/16678 or download from my site here: http://www.silent9.com/incoming/scripts ... colour.scm


It simply samples the entire image or active selection colour, and allows you to either:
- Display the RGB or grey values
- Set the foregound colour
- Set the background colour
- Add it to the current palette (if editable)

There is a toggle to sample just the layer, or the merged image.

It registers in the Colour menu as "Sample Average Colour..." and should work for RGB, indexed, or greyscale images.
Here is a quick pictorial explanation:


-Rob A>

Thanks Rob, I know (and use regularly) your script, very nice!
My "question" was about the Border Average filter, which IMHO is conceptually wrong, I don't understand why it has been kept as it is for so many years: it should have been rewritten with a more tightening logic.