Automatic 3D skull reconstruction using invariant features

Abstract

In this work we propose a new method to segment range images. It automatically extracts invariant features directly from point clouds. Points belonging to such features are used as the input to improve an evolutionary approach to 3D range image registration in forensic anthropology. We use such features in the automatic reconstruction of an accurate 3D model of the skull. Our reconstruction method includes a pre-alignment stage, that uses a subset of feature points, and a refinement stage. Results are presented over a set of instances of real problems.