[Insight-users] Problem related to the newly submited: Implementation of a 3D thinning algorithm

[Dan Mueller]

Hi Flo,

You may be interested in another Insight Journal article by Julien
Lamy regarding 3-D skeletonization:
    http://insight-journal.org/dspace/handle/1926/304
>From my understanding, this filter implements distance ordered
homotopic thinning for 2-D and 3-D images (cf Pudney). If you use the
filter, be warned that it currently can not handle images with
foreground pixels on the border...

I would also recommend that you write a review at the Insight Journal
for the filter you discussed. Doing so will allow the author of the
paper to be directly alerted to the issues, allowing them to respond
in an appropriate and traceable manner.

Cheers, Dan

On 16/10/2007, Flo <snrf at no-log.org> wrote:
> Dear all:
>
> I tried to use the newly submitted Implementation of a 3D thinning
> algorithm. However, using Colon images, I have wrong results.
> A bad behavior is illustrated here: http://florent.chandelier.free.fr/
> ITK/Thinning_3D_Colon_bad.bmp
> A good behavior is illustrated here: http://
> florent.chandelier.free.fr/ITK/Thinning_3D_Colon_good.bmp
> These are obtained on the same object but at different locations.
>
> As you can see, some portion of the skeleton are OK but some others
> are really messy (Zoomed Images). Indeed, I have no holes in my
> object but I end up with loops in the skeleton which thus modifies
> the Euler number.
>
> On a different note, I did myself work on Skeletonization during my
> PhD and *IF* I remember well, lee's paper (the referenced one) is not
> an exact method as pointed out by other papers (cf Pudney, 1997 I
> think. This is due to the cycle process while looking for the
> neighborhood. There is no unique cycle in 3D to go through the
> neighborhood as compared to the 2D case where there is a unique cycle
> - clock and counter-clockwise end up being the same.)
>
> This is a great effort to implement a 3D skeletonization process for
> ITK. Congrats to the ppl doing this. My personal view, and maybe a
> future contribution, would be to spend more time implementing the
> method described by Pudney, 1997 as it allows for the discrimination
> of every skeleton type (Exact medial surface, medial trunk and medial
> axis).
>
> I'm sorry to mail this to the news group but I do not know how the
> reviewing process works.
>
> regards,
>
> flo.