Harmonic Silhouette Matching for 3D Models

The ability to perform fast retrieval from a database of 3D models is becoming a growing necessity as the number of models in circulation is rapidly increasing. Several of the many existing methods dealing with this problem consider similarity measures based on visual appearance. This idea of comparing models using their respective silhouettes performs well on a number of benchmarks, but comes with a few inherent limitations, the biggest of which is that at the time of comparison all possible rotational alignments between two models need to be considered. In this paper we present two retrieval algorithms based on a silhouette representation. The first method shows how model similarity can be computed using fast harmonic matching techniques, and the second method reduces the problem to fast vector differencing using rotation-invariant properties of the representations.