Volumetric Density of Triangulated Range Images for Face Recognition

In this paper, a volumetric space representation of 3D range face image has been established for developing a robust 3D face recognition system. A volumetric space has been created on some distinct triangular regions of the 3D range face image. Further, we have constructed 3D voxels corresponding to those regions for developing voxelization-based 3D face classification system. The proposed 3D face recognition system has mainly three parts. At first, seven significant landmarks are detected on the face. Secondly, any three individual landmarks are used to create a triangular region; in this way, six distinct triangular areas have been generated, where the nose tip is a common landmark to all the triangles. Next, assume a plane at the nose tip level for representing the volumetric space. The total density volume and some statistical features are considered for the experiment. From the volumetric space, construct 3D voxel representation. Further, geometrical features from 3D voxel are used for the experiment. Three popular 3D face databases: Frav3D, Bosphorus, and Gavabdb are used as the input of the system. On these databases, the system acquires 94.28%, 95.3%, and 90.83% recognition rates using kNN and 95.59%, 96.37%, and 92.51% recognition rates using SVM classifier. Using geometric features with SVM classifier, the system acquires 92.09%, 93.67%, and 89.7% recognition rates.