Date: February 2020.
Source: International Journal of Computer Vision, vol. 128, pp. 547-571. DOI: 10.1007/s11263-019-01260-7.
Research Summary: This paper presents the extended, definitive version of the Liverpool-York Head Model (LYHM), built from the real-world 3dMD Headspace dataset of 1,212 subjects. The team introduced Iterative Coherent Point Drift (ICPD), a new correspondence algorithm combining Iterative Closest Points and Coherent Point Drift, paired with an adaptive template that personalizes the shape template to each subject’s facial features before dense morphing, and a Laplace-Beltrami Regularized Projection (LBRP) step to capture fine shape detail. The resulting model was aligned using Generalised Procrustes Analysis (GPA) and trained with Weighted Principal Component Analysis (WPCA), rather than standard PCA, to correct for uneven mesh resolution across the face and cranium. Benchmarked against the LSFM pipeline and the Basel Open Framework on the standard compactness, generalisation, and specificity metrics, the proposed pipeline outperformed both once more than 79 model components were used. A new high-resolution texture mapping technique, validated using SSIM, MS-SSIM, and IW-SSIM image quality metrics, also improved on standard per-vertex color mapping. In its clinical application, the model was used to evaluate 17 craniosynostosis patients across two surgical techniques, Barrel Staving (BS) and Total Calvarial Reconstruction (TCR), measuring outcomes by Mahalanobis distance from the population mean. Both techniques moved patients closer to the population norm, BS patients improving by 63.24 percent and TCR patients by 36.65 percent, described as the first use of full-head 3D morphable models in a craniofacial clinical study.
Conclusions:
Article: Statistical Modeling of Craniofacial Shape and Texture.
Authors: Hang Dai, Nick Pears, William Smith. Department of Computer Science, University of York. Christian Duncan, Alder Hey Children’s Hospital, Liverpool.