Content aware fill

Summary:

In Photoshop Cs5 there is a new feature called content aware fill. This is also aviable in Gimp called resynthesize. But it doesnt work well on edges. I have an idea for a script...

Content aware fill

Heho (guest) 20 Apr 2010 19:35

This script should do the following things
1. Find areas. This is my big problem. I want to find different areas based on there uhm whatever. For example a beach picture. After this first part the sky, the sand and the sea should be 3 different areas, but maybe also more sub divided(dark sand, light sand, deep sea you got it)
2. Find Area edges (which will be easy then)
3. Analyse edges where they touch the selected area(in which angle they touch them and which edges belong together)
4. create new areas(Based on the angle of the edges bezier curve shall be inserted. Which will create the new areas. This will be very complicated for mountains and so on)
5. look for similar areas and use them as patches for the first time resynthesize which will be done for every new area seperatly
6. Show the user the bezier curves and the areas from where the textures are taken. They should have the possibility to edit them
7. Resyntesize again and jump back to 6.

Re: Content aware fill

CarloB 21 Apr 2010 12:10

Actually, I can't see why you're posting this request here: this wiki is about the Liquid Rescale plugin, which is indeed distinct and very different from the Resyntesizer plugin.
Therefore, you should either contact the Resynthesizer author or maybe try the GIMP plug-in registry site (or any other site about extending GIMP functionality, provided it's more general than the present one).

Let me just note that dividing images into areas (also called image segmentation) is in general far from being a trivial task.

Also, the CA-fill feature is much probably based on something like patch-match (google for it if interested), rather then image segmentation.

Re: Content aware fill

Heho (guest) 21 Apr 2010 16:13

Yes but I didnt find something like contact…
I tried to contact the author but he won't answer.

yeah image segmentation seems to be the hardest part of it…
You are right. CA-fill is based on patch match and this is the part were my script should work better. Because with patch matching(which will also be used for the subdivided areas - this is the part of the resynthesize plugin) you get horrible artifacts on the edges of these areas. Once I figure out a good balance between much and sparse division the rest should be easy (nearly everything then is done by resynthesize).

Thanks for your answer