3D face recognition with asymptotic cones based principal curvatures
KAUST DepartmentApplied Mathematics and Computational Science Program
Computer, Electrical and Mathematical Sciences and Engineering (CEMSE) Division
Visual Computing Center (VCC)
Permanent link to this recordhttp://hdl.handle.net/10754/577097
MetadataShow full item record
AbstractThe classical curvatures of smooth surfaces (Gaussian, mean and principal curvatures) have been widely used in 3D face recognition (FR). However, facial surfaces resulting from 3D sensors are discrete meshes. In this paper, we present a general framework and define three principal curvatures on discrete surfaces for the purpose of 3D FR. These principal curvatures are derived from the construction of asymptotic cones associated to any Borel subset of the discrete surface. They describe the local geometry of the underlying mesh. First two of them correspond to the classical principal curvatures in the smooth case. We isolate the third principal curvature that carries out meaningful geometric shape information. The three principal curvatures in different Borel subsets scales give multi-scale local facial surface descriptors. We combine the proposed principal curvatures with the LNP-based facial descriptor and SRC for recognition. The identification and verification experiments demonstrate the practicability and accuracy of the third principal curvature and the fusion of multi-scale Borel subset descriptors on 3D face from FRGC v2.0.
CitationTang, Y., Sun, X., Huang, D., Morvan, J.-M., Wang, Y., & Chen, L. (2015). 3D face recognition with asymptotic cones based principal curvatures. 2015 International Conference on Biometrics (ICB). doi:10.1109/icb.2015.7139111