Date: May 2018.
Source: 2018 13th IEEE International Conference on Automatic Face and Gesture Recognition (FG 2018), pp. 91-97. DOI: 10.1109/FG.2018.00023.
Research Summary: Building on their earlier template morphing work, researchers at the University of York and Alder Hey Children’s Hospital developed Symmetry-aware Coherent Point Drift (SA-CPD), a shape morphing method for objects with approximate reflective symmetry, such as the human head. Standard morphing algorithms suffer from tangential sliding, where a template’s surface points drift sideways along the target shape rather than landing in their correct anatomical position. SA-CPD constrains all deformations to remain symmetric, using a Laplace-Beltrami regularized projection to fit any remaining asymmetries, directly addressing this problem. Evaluated against competing methods, including standard CPD, NICP, the LSFM pipeline, and the team’s own earlier CPD-LB approach, on 1,212 subjects from the real-world 3dMD Headspace dataset, SA-CPD outperformed all of them on both evaluation metrics: 87 percent of its Nearest Point Error (NPE) fell under 1mm, compared to 30 percent for CPD-LB and 28 percent for LSFM, and 99 percent of its Symmetry Contour Error (SCE) fell under 2mm, compared to 82 percent for CPD-LB and under 1 percent for LSFM. Using the resulting shape models to train Support Vector Machine (SVM) classifiers with 10-fold cross-validation, SA-CPD also achieved the best gender and age classification accuracy among all methods tested.
3dMD’s Role: Evaluated using 1,212 subjects from the real-world 3dMD Headspace dataset, which also underlies the Liverpool-York Head Model (LYHM).
Article: Symmetric Shape Morphing for 3D Face and Head Modelling.
Authors: Hang Dai, Nick Pears, William Smith. Department of Computer Science, University of York. Christian Duncan, Alder Hey Hospital, Liverpool.