Geometry Based Three Dimensional Face Expression Recognition

Abstract

This study discusses the problem of three-dimensional face expression recognition and presents a novel approach for recognizing facial expressions in the presence of three-dimensional geometry of face models. As a fundamental step in mesh processing pipeline using 3D scanned data, we first perform data alignment between different meshes. After registration step, we apply mesh smoothing algorithms which filter out high frequency noise and Taubin smoothing is used for regularization. Proposed approach is based on modeling surface geometry with completely regular triangle meshes called geometry images. Results demonstrating face expression recognition rates using end-to-end learning are presented and a three-dimensional facial expression database is used to provide a series of experiments.