The term surface extraction refers to extraction of a specific and meaningful surface from the data produced by some imaging device capable to produce three dimensional data. The data can be in form of unorganized point sets, range images or volumetric images. For example, many medical imaging devices produce volumetric images composed of voxels. Based on intensity values of voxels, one would like then to extract surfaces of anatomical or functional structures.
Since images are noisy and contain also other artifacts, automated surface extraction is not straight-forward. Bottom up approaches such as the Marching Cubes algorithm [5] suffer from their inability to consider information that is not present in the data. On the other hand, deformable models are approaches to surface extraction that can take also information extrinsic to image data into account. Particularly interesting for applications involving biological shapes - which are highly variable - are free form deformable models that make very few assumptions about the shape of the surface being extracted. Deformable surface meshes are instances of these. They can be used also to extract surfaces from range images, but here we consider our input to be a volumetric image. See Fig. 1 for an example of the surface extraction result from a noisy volumetric image using a deformable mesh.
Figure 1: a) The surface to be extracted from a noisy volumetric image, b) a
cross section of the noisy image, c) an extraction result using
a deformable mesh.