Segmentation and superquadric modeling of 3D objects

Abstract

A new model for representing an unorganised 3D data points set is proposed. Based on superquadrics, this model allows to describe the points set with a union of superellipsoids. Two different segmentation and modeling methods are developed in order to determine the whole model: a region growing approach and a split and merge one. This second method leads to a low sensitive model compared to the one obtained by the region growing. The model is simple and compact: only 11 parameters are needed per superellipsoid. It seems promising for 3D object compression and 3D object indexing and retrieval. As the topological relations of the superellipsoids are known, the model can be associated to a graph. The graph theory can thus be used in order to compare and to measure the similarity between 3D objects.