Segmenting 3d mesh images of the human face by local quadric parameterization

An approach to 3D surface (mesh) segmentation is proposed, consisting in clustering vertices by approximating quadric surfaces. Seed groups of adjacent points give rise to initial quadric patches, which then grow by absorbing unassigned mesh vertices and compete with neighbouring patches by seizing their vertices. While the algorithm is intended for general use, it has been applied to 3D representations of human faces and the quadric patches tend to settle on salient facial features such as cheeks, chin or the ridge of the nose. Early results are presented and further work outlined.