3D Facial Landmark Localisation by Matching Simple Descriptors

Abstract

We present our graph matching approach for 3D facial feature localisation. The work here uses a basic graph model (three vertices and three arcs) to locate the inner eye corners and the nose tip simultaneously. We intend to extend this to a larger set of the eleven features that exist in our ground truth of the face recognition grand challenge (FRGC) database. We apply the structural matching algorithm "relaxation by elimination" using a simple "distance to local plane" node property and a "Euclidean distance" arc property. After the graph matching process has eliminated unlikely candidates, the most likely feature combination (left eye, right eye and nose tip) is selected, by exhaustive search, as the minimum mahalanobis distance over a six dimensional space, corresponding to three node variables and three arc variables. Our results on the 3D FRGC database are presented and discussed.