Shape preserving surface reconstruction using locally anisotropic radial basis function interpolants

In this paper we deal with the problem of reconstructing surfaces from unorganized sets of points, while capturing the significant geometry details of the modelled surface, such as edges, flat regions and corners. This is obtained by exploiting the good approximation capabilities of the Radial Basis Functions (RBF), the local nature of the method proposed in [1], and introducing information on shape features and data anisotropies detected from the given surface points. The result is a shape preserving reconstruction, given by a weighted combination of locally anisotropic interpolants. For each local interpolant the anisotropy is obtained by replacing the Euclidean norm with a suitable metric which takes into account the local distribution of the points. Thus hyper-ellipsoid basis functions, named anisotropic RBFs, are defined. Results from the application of the method to the reconstruction of object surfaces in IR are presented, confirming the effectiveness of the approach.