add alternative thinning method (skeletonize)

[armadillu]

This adds alternative thin (skeletonize) implementation to the library.

The provided thin() doesn't always result the results I was expecting. For example, given this input:

The original thin() gets you this:

This new alternative implementation thin2() gets you this:

The implementation comes from here http://felix.abecassis.me/2011/09/opencv-morphological-skeleton/.

[roymacdonald]

Hi Oriol!
did you test your PR against #256?

The thinning algorithm has been broken for a long while. I will try to benchmark these two different implementations later.
all the best

[armadillu]

hmm no I never saw it, I wish I did bc it seems to be a better (althogh slower) algorythm.

[armadillu]

mine should not conflict with anything, so Im pretty sure they can live side by side. I think it may be handy to have different implementations that perform to better/worse results at slower/faster speeds, maybe gather them all into a single thin() method with an enum argument with the algorythm name?

[roymacdonald]

Hi. +1 for

maybe gather them all into a single thin() method with an enum argument with the algorythm name?

[roymacdonald]

what do you mean by better results? shouldn't these be the same, just slower or faster algos?

[armadillu]

there are different implementations of thining, I think its considered not deterministic. For example, the one I'm sending here ends up with lines that are not always 1-pixel thin as you can see in the 'i's and 't's.

[roymacdonald]

I see. I had the idea that it had to always finish with a one pixel line.

[kylemcdonald]

thanks, i like the idea of these all living side-by-side, since they are all useful in slightly different situations. i might change the names a little in the future.