This MDSTk extension package is a collection of libraries, binary modules,
and shell scripts for vector segmentation of 2D images, as well as
volumetric medical data.
A principle task of the segmentation is image partitioning into a set
of non-overlapping regions. This package provides image segmentation
technique based on the 2D/3D Delaunay triangulation.
Triangular/tetrahedral mesh is used to partition image data into
elements, while the mesh construction process respects significant
image edges. Therefore, surfaces of image regions are well described
and can be easily derived. In order to provide high quality meshes,
the method is based on techniques like the isotropic meshing and
the variational tetrahedral meshing.
Finally, every element of the mesh (e.g. tetrahedron) is characterized
by its feature vector detailing underlying image data. The library
provides several clustering methods to classify these feature vectors
into groups - image segments. Some traditional methods like region
growing and merging are also present. All these techniques are
modified, so that they work with meshes instead of raster image data.
For more information about the vector segmentation, see papers
enclosed in the 'doc/papers' directory in the package explaining
this techniquein more detail.