A Fully Three-Dimensional Method for Facial Reconstruction Based on Deformable Models

TLDR

The study aims to validate a 3D facial reconstruction method by comparing a reference skull/facial dataset with a ground‑truth dataset. Using CT‑derived 3D models of skulls and faces, the authors perform semi‑automatic registration, match crest lines with an iterative closest point algorithm, compute a global parametric transformation, and apply it to reconstruct the unknown face. The authors discuss the reliability and challenges of this novel reconstruction technique.

Abstract

Abstract Two facial models corresponding to two deceased subjects have been manually created and the two corresponding skulls have been dissected and skeletonized. These pairs of skull/facial data have been scanned with a CT scanner, and the computed geometric three-dimensional models of both skulls and facial tissue have been built. One set of skull/facial data will be used as a reference set whereas the second set is used as ground truth for validating our method. After a semi-automatic face-skull registration, we apply an original computing global parametric transformation T that turns the reference skull into the skull to be reconstructed. This algorithm is based upon salient lines of the skull called crest lines; more precisely the crest lines of the first skull are matched to the crest lines of the second skull by an iterative closest point algorithm. Then we apply this algorithm to the reference face to obtain the “unknown” face to be reconstructed. The reliability and difficulties of this original technique are then discussed.