Part-based registration for expression resistant 3D face recognition

This paper exhibits the performance drop in 3D face recognition in the presence of expression and proposes a new 3D face registration and recognition method based on facial regions. For fast alignment, a registration method based on iterative closest point (ICP) algorithm is used which employs average regional models (ARMs). Similarity scores obtained from alignment with separate regional models are fused to increase the recognition performance. In this work, a new 3D face database is presented which is collected for studies on expression analysis and on face recognition in the presence of facial expression. The experimental results are reported on this new database.