Dense and Sparse 3D Deformation Signatures for 3D Dynamic Face Recognition

This work analyses dense and sparse 3D Deformation Signatures to represent temporal deformation instances. The signatures are employed in dynamic face recognition, however, they applicable other domains. is demonstrated for expression recognition. pushing need non-intrusive bio-metric measurements made its expressions recognition dominant players domains like entertainment, surveillance security. proposed signature can be computed from 2D, or hybrid input by means of robust fitting. It given a non-linear 6D space representation which guarantees construction physically plausible deformations. A unique indicator per triangle triangulated mesh as ratio derived scale in-plane the canonical space. These indicators concatenated densely sparsely form signature. then used learn facial signals. Two datasets were examined evaluation. reported 1-Rank accuracy outperforms existing literature. Democratising step results 100% confusion matrices. In an open-world setting context, was achieved detecting intruders. robustness has been further validated very challenging highly dataset.