3-D facial expression representation using statistical shape models

Abstract

This poster describes a methodology for facial expressions representation, using 3-D/4-D data, based on the statistical shape modelling technology. The proposed method uses a shape space vector to model surface deformations, and a modified iterative closest point (ICP) method to calculate the point correspondence between each surface. The shape space vector is constructed using principal component analysis (PCA) computed for typical surfaces represented in a training data set. It is shown that the calculated shape space vector can be used as a significant feature for subsequent facial expression classification. Comprehensive 3-D/4-D face data sets have been used for building the deformation models and for testing, which include 3-D synthetic data generated from FaceGen Modeller® software, 3-D facial expression data caputed by a static 3-D scanner in the BU-3DFE database and 3-D video sequences captured at the ADSIP research centre using a 3dMD® dynamic 3-D scanner.