a tube, or not a tube: that is the question

Hi all,
I got a 3d binary image, with one big structure connected.
This big structure is formed by different sub-structures (for
simplicity lets say: balls, tubes, ellipses, torus, and so on).
And I want to retrieve the original structures.
Unfortunately I do not know the size of the structures, but I know
(more or less) the proportion: a tube is bigger then another of a
factor 2/3.
Let's simplify: actually I need only the tubes ;-)
knowing the minimum and maximum size would be easy (smoothing with a
kernel of size greater then maximum and subtracting from the
original), but I don't know the sizes.
I was thinking to work with eigenvectors, but also with eigenvectors
this is a (reasonably) easy problem if the size of the tubes is (less
or more) known. Am I right? But the size is actually unknown...
well... it is in a range, but the range is quite wide (from 2 voxels
to 40, it doesn't help at all!!)
Any suggestion?

cheers,
Mauro


.

Quantcast