Relevance analysis of 3D curvature-based shape descriptors on interest points of the face

In this work, the behavior of six curvature-based shape descriptors k1, k2, Mean, Gaussian, Shape Index, and Curvedness computed at different locations of the face surface was evaluated over synthetic face models and 3D face range images data sets in order to establish the one that offers better discriminancy in different cases. A set of points selected from relevant parts of the human face was extracted. For evaluating the six descriptors over the selected points, the Fisher coefficient was used. Two kinds of tests were designed; the first one, to establish which descriptor is the most representative over all the set of points (global relevance); the second test was performed with sub sets of points from selected regions (local relevance). Finally, we obtain which descriptors have the most relevant information in the selected points of the face surface, which is an important step before performing a classification or recognition process.