Face recognition using 3d surface-extracted descriptors

The discriminating power of three dimensional (3D) descriptors extracted from 3D human face surfaces is analyzed. An automatic face recognition system using different subsets of the descriptor set has been implemented and tested. We used 420 3D-facial meshes belonging to 60 individuals, including views presenting light rotations and facial expressions, for the experiments. An HK segmentation (based in the signs of the mean and Gaussian curvatures) for isolating regions of pronounced curvature has been performed. Eighty six descriptors have been obtained from the segmented regions. The thirty five more discriminating ones in frontal views provide 78 % of recognition success rate when best match is selected, and a 92 % of recognition success is obtained when the five best matches are selected