[Insight-users] Multilabel dilation

[Richard Beare]

I think there are a number of options. It looks like you are after a
voronoi tesselation masked by your dilation radius.

A grayscale dilation applied to label maps will result in higher
labels overwriting lower ones, rather than splitting at the midpoint.

The MaurerDistanceTransformImageFilter has an option to output
tessellations, which may be usable for your application - I haven't
tried this. The strategy would be to use a binary dilation and use the
result to mask the tessellation, if the tessellation is a label image.

My traditional way of doing a label dilation (because I had forgotten
about the above) is to compute a distance transform of the binary
image, and a dilation of the binary image. I then use the original
label image as seeds for a watershed transform, and the distance
transform as the control surface. The result of the watershed is the
tessellation represented as a label image. You then mask with the
dilated binary. This can be done with filters in the standard ITK
distribution.

As the above is a bit messy, and label dilation is a task I've needed
quite often, I've developed an all-in one solution that uses parabolic
morphology operations that is much faster. I haven't published the
code for it yet though. If there is more interest I'll try to get my
act together.

On Fri, Feb 8, 2013 at 3:59 AM, Dženan Zukić <dzenanz at gmail.com> wrote:
> I have an image and n points (n small, 5-20). I want to calculate influence
> area of each point. Each voxel in an image can be influenced by at most 1
> point, by closest point if distance to it is less then let's say 50mm,
> otherwise it is not influenced by any point.
>
> I thought I could do this by clearing the image, labeling the n points with
> numbers 1-n, and then using DilateImageFilter, if it can somehow handle this
> case. Perhaps do it in steps: dilate by a small amount (5mm). See attached
> images.
>
> Is there something like multi-label dilation image filter? Or does
> gray-scale dilation filter does this? How about PadLabelMapFilter? Or would
> it even be faster and more convenient to iterate through all the voxels,
> calculate distances to all n points and then deduce to which point the given
> voxel belongs, if any?
>
> _____________________________________
> Powered by www.kitware.com
>
> Visit other Kitware open-source projects at
> http://www.kitware.com/opensource/opensource.html
>
> Kitware offers ITK Training Courses, for more information visit:
> http://www.kitware.com/products/protraining.php
>
> Please keep messages on-topic and check the ITK FAQ at:
> http://www.itk.org/Wiki/ITK_FAQ
>
> Follow this link to subscribe/unsubscribe:
> http://www.itk.org/mailman/listinfo/insight-users
>