A method based on geometric invariant feature for 3D face recognition

3D information provides a significant improvement in recognition performance over 2D facial image data. However, the existing 3D approaches show limitations dealing with pose variation, e.g., 3D facial surfaces need to be aligned before the match operation. In this paper, an original framework which has the scale, rotation and expression invariance based on geometric invariant feature is proposed for automatic face recognition without pre-registration. In this study, 3D face scans are first pre-processed, including mesh cropping, holes filling, and mesh regularization; subsequently, the geometric invariant feature combined the local shape variation feature with spatial geometric feature which is invariant to scale and pose is extracted. Experimental results implemented on GavabDB and our purpose-selected database demonstrate that our proposed method significantly outperforms the state-of-the-art methods with respect to pose and facial expression variation.