AW: AW: [Insight-users] Question about Conntected Regions

[Luis Ibanez]

Hi Salah,

I will insist in going back to your segmentation process.

There is always more information before segmentation than
after.  Concepts that are quite obvious for the human eye
may result pretty hard to grasp with image processing
algorithms. That is, you may be attacking the problem in
a point where it is not easy to solve.

It will be easier to figure out a way of differentiating
both regions in the original image before segmentation.

If the figures have well defined shapes, you could attempt
to achieve segmentation by doing model-to-image registration,
using one model for each shape. This will be much more
efficient.

Could you give us some details about what this segmentation
is intended for ?  e.g. anatomical structure, image modality...



    Luis


------------------------

salah wrote:
> Hi Luis,
> 
> Unfortunately, the original image does help distiguishing
> between the wanted and un-wanted regions. This is why I am
> trying to fix the segmented output image.
> 
> A friend has proposed me the following: 
> " Choose two seed points, one in each of the regions. 
> Then find the shortest path between them. (This is actualy
> related to my other email concerning the shortest path 
> question) and then cut the two region at narrowest region
> surrounding this path."
> 
> I think, this solution would not be feasible. Finding the 
> shortest path would be a fairly time-consuming step! 
> Is that true ?? Moreover, finding the narrowest region around
> the path would also be not that easy!!
> By the way, I looked at the Fuzzy Connectedness Filter as 
> you suggested. This filter computes the strongest path between 
> very voxel in the image and the seed voxel. It really does NOT 
> give out or maintain this path, does it?? Knowing the strength 
> of a path does not help finding it!! Right??
> 
> Please tell me if you had some other idea.....
> 
> Many thanks,
> Zein
> 
> 
>