3D modelling and segmentation with discrete curvatures

Recent concepts of discrete curvatures are very important for Medical and Computer Aided Geometric Design applications. A first reason is the opportunity to handle a discretisation of a continuous object, with a free choice of the discretisation. A second and most important reason is the possibility to define second-order estimators for discrete objects in order to estimate local shapes and manipulate discrete objects. There is an increasing need to handle polyhedral objects and clouds of points for which only a discrete approach makes sense. These sets of points, once structured (in general meshed with simplexes for surfaces or volumes), can be analysed using these second-order estimators. After a general presentation of the problem, a first approach based on angular defect, is studied. Then, a local approximation approach (mostly by quadrics) is presented. Different possible applications of these techniques are suggested, including the analysis of 2D or 3D images, decimation, segmentation... We finally emphasise different artefacts encountered in the discrete case.