Segmenting contours

zetah ven otrov at hush.ai
Sun Feb 24 23:49:37 EST 2013


Hi Stefan, and thank you for your reply

For some reason I did not receive email notification, and came here to post 
what I found, when I was pleasantly surprised by your reply :)

I was looking through OpenCV functions just yesterday, and I didn't notice 
this Inpaint() magic function, as I was expecting something in other parts 
of this library, like segmentation, tracking and countouring. Contouring 
ringed the most, but it was complicated for me at the time to extract 
something meaningful for my problem which is topological contour tracing 
(that I colored with above example images, not to scare potential reader ;)

I just tried the function, but I guess it works best for photo images, as 
line contours aren't reconstructed as expected. Anyhow thanks for the tip, 
it could be handy in future occasions I guess.

I was also reading various papers. There seem to be couple of methods for 
reconstructing dense contours with gaps, which is part of more general 
problem of curve reconstruction. The most accurate results are reported by 
using Delaunay triangulation and properties of Voronoi diagram vertices, 
and then applying filtering to obtain the reconstruction. It's referenced 
to Amenta paper "The crust and the beta-skeleton: Combinatorial curve 
reconstruction" and it seems to me that environment like NetworkX could be 
my first step in trying to visualize and then understand this method, as 
although I've heard about Voronoi and Delaunay triangulation, I never used 
those functions. Then perhaps engage scipy and matplotlib for real example.
In slow pace that is my plan.

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.python.org/pipermail/scikit-image/attachments/20130224/a7baa3dd/attachment.html>


More information about the scikit-image mailing list