Interactive local image masking algorithm using points as reference

I remember in one of the treads discussing U points with @anon41087856 - method of interactively masking a specific image area of the image using reference points (as used by proprietary editors like Viveza DxO or Snapseed).

Recently I found an article about similar approach which might be of interest for some developers here:

Website:

http://chiakailiang.org/touchtone/

and

PDF paper:

http://chiakailiang.org/touchtone/touchtone.pdf

1 Like

Should be possible with G’MIC as it already has selection tools. @David_Tschumperle

Here I was thinking you need a computer to do something like what @afre has shown. But, it’s still limited in scope though, but it’ll do for most people.

My ‘local similarity mask’ in the GMIC GIMP plugin does a similar thing for creating a mask.

A very nice filter indeed @Iain!

I use it sometimes and in some cases it works very well.

But,…
(In this example I try to mask the areas with red autumn leaves on the floor)

  1. Sometimes there are issues with the preview. This is what the selection in G’MIC looks like:

And after the filter is applied in GIMP:

  1. it would be nice if you could have several reference points to combine them similar to interactive extract foreground filter from @David_Tschumperle to make the mask more accurate

  2. that the influence of the points can be limited regionally (to be able to modify the effective radius)

  3. that you can switch between mask and image view (to be able to see exactly what has been masked like in the extract foreground filter mentioned above):

That’s a universal problem that exists when the preview is a different scale than the final image. For some filters it doesn’t matter, but for others it’s a problem. There’s not much I can do I’m afraid.

I have been playing with having a ‘negative’ point, for areas you want to exclude, but I have not been happy with it. I think having multiple points would be difficult because for this filter each point would need its own mask. That would become to slow or memory intensive.

I’ve been trying this as well, but once again, nothing satisfactory.

That’s a bit easier to accomplish, I’ll put that on my to-do list.

1 Like

Thank you so much for your feedback and effort @Iain!
I really appreciate your work!

Perhaps some of the solutions described in the PDF document above could help to achieve some of goals.