Date: September 2025.
Source: The Visual Computer, 2025, 41(11), pp. 9077-9090. DOI: 10.1007/s00371-025-03914-9.
Research Summary: Researchers at Beihang University introduced Edge-Constrained Non-Rigid ICP (Edge-NICP), a fully automated method for aligning a 3D face template to a target scan, a foundational step for 3D face reconstruction, facial expression analysis, and facial recognition. The method initializes registration using a 3D Morphable Model for robust starting conditions, then refines correspondence using edge length constraints to keep the mesh’s structure consistent as it deforms, addressing a known weakness where standard non-rigid ICP can distort mesh geometry during alignment. A new global spatial similarity metric evaluates how consistently the registered mesh’s vertices land in the correct position across different subjects. Tested on the real-world 3dMD Headspace dataset, the method achieved an average spatial similarity of 98.23 percent, outperforming existing techniques.
3dMD’s Role: Headspace is the real-world 3dMD dataset, 1,519 volunteer subjects, captured by Alder Hey Children’s Hospital in a Liverpool City Centre public forum from 26 September to 1 December 2013. This paper’s authors accessed it as a publicly available research resource. Headspace served as this paper’s evaluation benchmark for testing registration accuracy.
Article: Enhanced 3D Facial Registration via Edge-Constrained Non-Rigid ICP and Global Spatial Similarity Evaluation.
Authors: Yaopu Zhao, Guanghong Gong, Ni Li, Yanchao Yu. School of Automation Science and Electrical Engineering, Beihang University, Beijing, China. Ying Li, School of Police Information, Shandong Police College, Jinan, China.