On discriminative properties of TPS warping parameters for 3D face recognition

Due to the advances in the acquisition systems, three-dimensional (3D) facial shape information has been increasingly used for human face recognition. In order to be able to compare different facial surfaces, various registration approaches have been proposed, including Thin Plate Spline (TPS) based algorithms. In this paper, instead of adopting the TPS for registration purposes, we analyze the discriminative properties of the parameters obtained by deforming a generic face model onto target faces using TPS. The warping parameters (WP) that describe the non-global and non-linear transformations and represent the deviations from the common geometric structure are given to the classifier for face recognition. The descriptiveness of those vectors is analyzed on the FRGC database where total of 4569 3D face models are utilized. In spite of its low complexity compared to other proposed approaches, this method yields promising accuracy rates.