Authors:
Zahra Toony
;
Denis Laurendeau
;
Philippe Giguère
and
Christian Gagné
Affiliation:
Université Laval, Canada
Keyword(s):
3D segmentation, Spectral 3D segmentation, Primitive extraction from mesh models, Clustering.
Related
Ontology
Subjects/Areas/Topics:
Computer Vision, Visualization and Computer Graphics
;
Fundamental Methods and Algorithms
;
Geometry and Modeling
;
Modeling and Algorithms
;
Scene and Object Modeling
;
Surface Modeling
Abstract:
Being able to automatically segment 3D models into meaningful parts is an important goal in 3D shape processing.
In this paper, we are proposing a fast and easy-to-implement 3D segmentation approach, which is
based on spectral clustering. For this purpose, we define an improved formulation of the similarity matrix
which allows our algorithm to segment both free-form and CAD (Computer Aided Design) 3D models. In 3D
space, different shapes, such as planes and cylinders, have different surface normal distributions. We defined
the similarity of vertices based on their normals which can segment a 3D model into its geometric features.
Results show the effectiveness and robustness of our method in segmenting a wide range of 3D models. Even
in the case of complex models, our method results in meaningful segmentations. We tested our segmentation
approach on real data segmentation, in the presence of noise and also in comparison with other methods which
provided good results in all cases.