[Insight-users] MPI Fast Marching

[Kevin H. Hobbs]

On 10/12/2009 06:52 PM, Luis Ibanez wrote:
> Hi Kevin,
>
> Thanks for bringing this document to our attention.
>
> Before we read the 130 pages...   :-)
>   

I basically just skipped to the parallel part.

> Do you have a specific question about it ?
>
>   

Yes actually.

 Tugurlan describes a way to restart fast marching on each piece.

If I understand the dissertation correctly, then after each iteration
she finds the minimum fast marching arrival time from all of the ghost
points, of all of the neighbor pieces, and relabels all of the points in
the piece as either accepted if their time is less than that minimum
time, narrow band if their time is equal to the minimum time, or far
away if their time is greater than the minimum time.

That seems reasonable to me.

 The only thing I worry about is "equal."

It seems straight forward enough to set the alive and trial points in an
iterator after each parallel itteration.

But if the pixel type is float or double than no pixel will ever be
equal to another.

Picking some tolerance for equal is probably silly.

Contouring the image and setting the new trial points as those on the
contour might work.

What do y'all think?

-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 260 bytes
Desc: OpenPGP digital signature
URL: <http://www.itk.org/pipermail/insight-users/attachments/20091013/95079b78/attachment.pgp>