Characteristics measurement for three-dimensional surfaces based on multiple filtering

Matching a 3-D shape to a model in a database is one of the main issues in 3-D computer vision, along with indexing 3-D objects for the teleconferencing with realistic sensations. Concerning this issue, one of the keys for reducing the error and computational cost in the matching step is to carry out robust characteristics measurement of the 3-D shape. This paper proposes a new multiple spatial filtering based characteristics measurement technique for 3-D surfaces, and presents experiments based on qualitative and quantitative analyses comparing characteristics obtained by the technique and conventional differential geometric characteristics. This method uses the trajectories of a point on a surface caused by viewpoint invariant filtering. By using a set of low-pass filters with multiple radii, the shape of the trajectory reflects some shape information around the point on the surface, from coarse to fine, while eliminating noise. Basic concepts, the characteristics measurement technique, and experimental results will be shown.