[Numpy-discussion] Smoothing binary volumes

Thiago Franco Moraes totonixsame at gmail.com
Tue Sep 6 14:15:46 EDT 2011

Hi all,

I've been implementing the algorithm from this paper "Reducing
Aliasing Artifacts in Iso-Surfaces of Binary Volumes" from Ross T.
Whitaker. Because I develop a opensource software which works with
segmentation of CT and MRI medical images, and the results of
segmentation is a binary volume image. The problem is that the surface
generate from binary images are rough and have those staircase
artifacts. Bellow some examples after and before using the algorithm:

Binary ball - http://i.imgur.com/DlIwP.png
Smoothed ball - http://i.imgur.com/G04zN.png

Binary CT - http://i.imgur.com/Ah1LB.png
Smoothed CT - http://i.imgur.com/ps1Nz.png

The algorithm works in the volumetric image. After the algorithm I use
the Marching Cubes (from VTK) to generate the surface.

I could use a gaussian filter, but some fine features could be lost.

The source-code is here [1]. The first attempt I use python, but it
used a lot of memory. Now I'm using Cython, it's not completed. Maybe
I will try to code directly in C.

I'm sending this email to this mail-list because someone could have
some interest in this, and maybe it could be inserted in some python
library, like mahotas.


[1] - https://github.com/tfmoraes/binvol

More information about the NumPy-Discussion mailing list