Date: November 2017.
Source: Journal of Imaging 2017, 3(4), 55; doi:10.3390/jimaging3040055.
Research Summary: Extending their earlier single-viewpoint work, researchers at the University of York and Alder Hey Children’s Hospital built a fully automatic pipeline to construct 2D morphable models of craniofacial profile from three orthogonal viewpoints, side, front, and top, using 3dMD head scans from the real-world 3dMD Headspace dataset, subjects wearing a close-fitting latex cap to reveal skull shape. Texture-based 3D pose normalization and facial landmarking extract the profiles from each raw scan, which are then brought into dense correspondence through automatic annotation, subdivision, and registration, scaled and aligned using Generalised Procrustes Analysis (GPA), before applying Principal Component Analysis (PCA) to build the morphable model. The team also introduced a new alignment method, Ellipse Centre Nasion (ECN), as an alternative to GPA. Validated in a craniosynostosis intervention outcome case study, the model achieved state-of-the-art results. Both the morphable models and the underlying profile dataset were made publicly available.
Article: Modelling of Orthogonal Craniofacial Profiles.
Authors: Hang Dai, Nick Pears, and Christian Duncan. Alder Hey Children’s Hospital, Liverpool and Department of Computer Science, University of York, York, United Kingdom.