A 3D Face Recognition System for Expression and Occlusion Invariance

Facial expression variations and occlusions complicate the task of identifying persons from their 3D facial scans. We propose a new 3D face registration and recognition method based on local facial regions that is able to provide better accuracy in the presence of expression variations and facial occlusions. Proposed fast and flexible alignment method uses average regional models (ARMs), where local correspondences are inferred by the iterative closest point (ICP) algorithm. Dissimilarity scores obtained from local regional matchers are fused to robustly identify probe subjects. In this work, a multi-expression 3D face database, Bosphorus 3D face database, that contains significant amount of different expression types and realistic facial occlusion is used for identification experiments. The experimental results on this challenging database demonstrate that the proposed system improves the performance of the standard ICP-based holistic approach (71.39%) by obtaining 95.87% identification rate in the case of expression variations. When facial occlusions are present, the performance gain is even better. Identification rate improves from 47.05% to 94.12%.