[computer-go] finding groups as an image segmentation problem

["Robin Kramer" <robin@xxxxxxxx>]

Hi,

It struck me today that finding groups on the go board is analogous to the
image segementation problem from the field of computer vision.

There are several algorithms which have been used to segment images into
useful segments.

The one that I think might apply to Go is the "Segmentation by
graph-theoretic clustering"

the idea being that one would construct an affinity matrix that is 361X361,
each element in the matrix would represent how much affinity each position
had to each other position.  There is an algorithm based on eigen vectors
which will find the cuts which maximize the affinity of the ellements in
the same cut/group.

The problem in using this method is how does one measure the affinity?  I
would think stones of the same color, on the same chain and in the same
neighbor hood would have a high affinity.

Has anyone tried using image segmentation algorithms before?  Are their
potentially other good affinity measures?  I am working from a text book
now, by Forsyth and Ponce if any one is interested I can try to find a
paper or send references.

Sincerely,

Robin Kramer


_______________________________________________
computer-go mailing list
computer-go@xxxxxxxxxxxxxxxxx
http://computer-go.org/mailman/listinfo/computer-go