On Jun 5, 3:44=A0pm, ImageAnalyst <imageanal...@[EMAIL PROTECTED]
> wrote:
> On Jun 5, 8:59=A0am, luca.pampar...@[EMAIL PROTECTED]
wrote:
>
> > Hi everyone,
>
> > I am currently doing a threhold based segmentation (Otsu segmentation)
> > on an image and would like to connect some of the voxels using
> > dilation. What is the normal approach here? Usually, what kind of
> > structuring element would one use? is there a rule to determine this
> > or do you just try different ones till one of them seems to work?
> > Also, what about the subsequent erosion? What sort of structuring
> > element is recommended there?
>
> > Thanks,
> > Luca
>
> ----------------------------------------------------
> Luca:
> I usually do a closing (dilation followed by erosion) rather than a
> dilation because I want to end up with the same size structures before
> segmentation rather than larger structures. =A0I use a round structuring
> element because if I use a box or cube I get artifacts. =A0If you use a
> square structuring element on an image of a thin ring-shaped
> structure, you'll see bogus artifacts at 45 degrees, 135 degrees,
> etc. =A0These are minimized with a circle. =A0Try different window sizes
> and on the original gray scale image or on the binarizy, thresholded
> image and just see what works best in your situation.
> Regards,
> ImageAnalyst
As said already,
[1] The scale of the structuring element can be determined by the size
of the smallest opening you want to keep open after the 'closing'.
[2] The shape of the structuring element can be determined by the
shape of the smaller features in the image.
Prashant Chopra
Moderator,
Science.Medical.Imaging group
http://groups.google.com/group/medicalimagingscience
|
