3D face recognition using pose invariant nose region detector

The 3D faces matching a challenging problem lying at the core of essential research areas in 3d face recognition systems and requires a crucial registration technique in order to achieve high matching rates. In this work, we address within a new framework, the keypoints detection problem and face registration in 3D partial facial scans (implies missing parts). The major contribution of this work is to improve robust matching approach against wide pose variations (up to 60° around Yaw axis) using nose region extracted. A construction of average nose model is conducted for alignment purpose. The facial representation is formulated using local shape descriptors extracted from surface keypoints of interest regions.. Faces similarity is performed by applying first Iterative closest point (ICP) method between each nose subject in the gallery and the probe one, and thus takes the advantageous of minimum alignment error. Second, select minimum alignment errors that related to the first five subjects, and then manage a local shape descriptors comparison to determine matching score. The effectiveness of the proposed approach has been evaluated on GAVADB 3D face database which consists of both frontal and partial facial scans. The proposed approach achieved recognition rate about 94% and 90% for frontal and partial facial scans.