Robust alignment of salient facial regions for recognition of 3-D partial faces scans

The work in this paper is dedicated to present and experiment a fully automatic face recognition approach based on exploiting the distinctive traits of 3D facial scans. We aim to present a recognition approach operates with fully and partial facial scans (missing facial parts). A region based approach for registration and recognition processes is adopted to offer robust faces matching against facial expression and pose variations. An average nose model was constructed in this work via procrustes analysis concept for registration purpose. The nose region for both training and testing facial scans is detected using cascade filtering scheme of the extracted local descriptors (Distance to Local Plan and shape index). Finally, the similarity measure between faces is computed based on Iterative Closest Point (ICP) method. The effectiveness of the proposed approach has been evaluated on GAVADB 3D face database which consists of both frontal and partial facial scans. The experimental results showed, that nose region has been detected accurately with success rate 98% for facial scans having natural expression and frontal pose, which leads to achieve high recognition rates of the faces. The experiments have demonstrate that nose region based detector capable to operates significantly with partial facial scans and thus achieves accuracy about (5.5 mm) for nose tip location.