Segmentation and Volume Representation Based on Spheres for Non-rigid Registration

Abstract

AbstractThis paper presents three different tasks: segmentation of medical images, volume representation and non-rigid registration. The first task is a necessary step before volume representation ant it is done with a simple but effective strategy using tomographic images, combining texture and boundary information in a region growing strategy, obtaining good results. For the second task, we present a new approach to model 2D surfaces and 3D volumetric data based on marching cubes idea using however spheres (modeling the surface of an object using spheres allows us to reduce the number of primitives representing it and to benefit -from such reduction- the registration process of two objects). We compare our approach based on marching cubes idea with other one using Delaunay tetrahedrization, and the results show that our proposed approach reduces considerably the number of spheres. Finally, we show how to do non-rigid registration of two volumetric data represented as sets of spheres using 5-dimensional vectors in conformal geometric algebra.KeywordsSeed PointIterate Close PointGeometric AlgebraVolumetric DataVolume RepresentationThese keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.