Enhanced 3D facial registration via edge-constrained non-rigid ICP and global spatial similarity evaluation

Three-dimensional (3D) facial registration is critical for accurate facial reconstruction and expression analysis, aligning diverse data across sources and time intervals. In this paper, we introduce an enhanced 3D facial registration algorithm that leverages edge-constrained non-rigid iterative closest point (ICP) and a global spatial similarity metric. Our approach initializes the registration process using a 3D Morphable Model, which provides robust initial conditions. We propose an edge-constrained non-rigid ICP algorithm to refine the correspondence and non-rigid transformation between shapes. This algorithm incorporates edge length constraints to maintain mesh structure consistency during deformation. A global spatial similarity metric is employed to evaluate the consistency of vertex positions across models, ensuring accurate facial registration. Experiments on the Headspace dataset demonstrate that our method achieves an average spatial similarity of 98.23%, outperforming existing techniques. Our fully automated framework provides a reliable foundation for applications such as 3D face reconstruction, facial expression analysis, and facial recognition, setting the stage for future advancements in the field. The code for our algorithm is open-sourced and available at https://github.com/zhaopu99/edge-nicp.