Nonrigid-deformation recovery for 3D face recognition using multiscale registration.

Nonrigid deformation is a fundamental feature in face recognition. The proposed method extracts nonrigid deformation by finding the mapping between two shapes. To improve registration accuracy, it integrates geometric shape decomposition and nonrigid point-set registration. On the basis of manifold harmonics, the method first decomposes shapes into low-frequency and high-frequency parts. Then, it applies the modified registration algorithm to obtain deformation parameters. An extensive set of experiments evaluated the method's performance on the FRGC (Face Recognition Grand Challenge) v2 database. The method achieved not only higher accuracy for fitting but also 98.2 percent identification and 97.4 percent verification at a 0.001 false-acceptance rate.