This is really cool! I've seen other techniques trying to solve the same problem, such as randomly rotating each tile, but this looks much cleaner and there are no "breaks" at the edges. How computationally expensive is this? Would it be feasible to do it real time (i.e. to create forest tiles for a game)?
kkkhanl
Hmmm, I think it is interesting for the second pattern for example that the 5x5 looks so different from the given texture. From a person's perspective it seems like the pattern can be easily identified, I think it is interesting that it is much more difficult for the computer to do this job.
sartk
As the search window size approaches infinity, does the result keep getting better? Or does it approach just an exact tiling of the image, which will lead to jagged connections.
joeyhou0804
I wonder how will these results change when the size of the neighborhood search window keeps increasing.
This is really cool! I've seen other techniques trying to solve the same problem, such as randomly rotating each tile, but this looks much cleaner and there are no "breaks" at the edges. How computationally expensive is this? Would it be feasible to do it real time (i.e. to create forest tiles for a game)?
Hmmm, I think it is interesting for the second pattern for example that the 5x5 looks so different from the given texture. From a person's perspective it seems like the pattern can be easily identified, I think it is interesting that it is much more difficult for the computer to do this job.
As the search window size approaches infinity, does the result keep getting better? Or does it approach just an exact tiling of the image, which will lead to jagged connections.
I wonder how will these results change when the size of the neighborhood search window keeps increasing.