Abstract
We investigate the use of Hermite Radial Basis Functions (HRBF) Implicits with least squares for the implicit surface reconstruction of scattered first-order Hermitian data. Instead of interpolating all pairs of point-normals, we select a small subset of point-normals as centers of the HRBF Implicits while considering all pairs as least-squares constraints. Centers are adaptively sampled via a novel greedy algorithm that takes into account Hermitian data and distances between points. This approach produces sets of centers that are globally well distributed and preserves local features. We show that this yields accurate surface reconstructions with small sets of centers.
