[Insight-users] surface reconstruction based on membership images

[Bruno Arbter]

For surface reconstruction in a 3D grey-scaled data set I am using marching
cubes. In order to have a semi-automatic procedure for finding reasonable
isovalues, I had a look at the classification methods described in the
SoftwareGuide (e.g itkBayesianClassifierImageFilter). The membership
function images from itkBayesianClassifierImageFilter fits quite well to the
expected membership.

Using the resulting classified binary images as input for the surface
reconstruction I naturally get aliasing. Now my intend was, to use the
membership images (NOT the final binary images!) as an input for surface
reconstruction. But when doing marching cubes on the membership images I
again got slight aliasing (or similar effects). Using the original data set
there is no aliasing.
 
Is there an theoretical explanation for this effect?

Thanks
Bruno

P.S.: I tried to send an attachment, but the message with attachment didn't
show up in the list...