3D face registration by depth-based template matching and active appearance model

Automatic 3D face registration is highly important for 3D face recognition, which can also be used in facial feature segmentation, facial mesh reconstruction, face synthesis and motion capture. In this paper, we propose a coarse-to-fine 3D face registration approach based on template matching of depth images and depth-based active appearance model (AAM). First we construct three multi-angle nose templates to detect nose regions even in situations such as missing of facial data in which the symmetrical property of face is destroyed. The Normalized Cross Correlation (NCC) is utilized as the template matching method to obtain an estimated pose of the faces and then roughly align the faces to near-frontal views. A depth-based AAM model is then built to finely align the faces. Through this coarse-to-fine registration procedure, our method is robust against pose and expression variations. Experimental results indicate that the proposed method outperforms the previously proposed landmark detection method and achieves good performance on the Bosphorus 3D face database.