Triangle meshes have become a popular and convenient piecewise linear
approximate representation of surfaces. It is desirable to segment a triangle
mesh into its parts in order to get some meaningful structural description of
the shape. This is useful in many applications. Few automatic feature
extraction techniques exist currently.
This paper presents a technique for finding feature lines: piecewise
linear curves on the triangle mesh separating the mesh into segments. These
lines approximately pass along the ridges of maximum inflection (maximum
curvature change). The basic method is to compute curvature values at every
vertex of the mesh, and then threshold the values to create a binary
feature vector, where each element of the vector represents the binary
value assigned to a particular vertex.
Then noise and artifacts can be removed from the features using
morphological operators. Finally the paper presents an operator that extracts
a skeleton of the mesh, which is a representation of the structure of the
shape.
Usually multiple steps are needed to find all of the desired patch boundaries.
The method in this paper is one of the first automatic feature extraction
algorithms and may be applied to classifying surfaces sampled from laser range
scanners. The definition of the neighborhood of a vertex must be changed for
reduced meshes with triangles of strongly varying edge length. Also, manual
assistance is still needed for picking thresholds or fixing the feature
vector.