Hierarchical Evaluation Model: Extended Analysis for 3D Face Recognition

In this paper we present a method for 3D face recognition that is suitable for verification systems. A Simulated Annealing (SA)-based approach for range image registration is used to perform 3D face matching. The Surface Interpenetration Measure (SIM) is used during the registration process to assess precise alignments. This measure is then used as similarity score between two face images. In the verification scenario, we propose a hierarchical evaluation model to answer if two face images belong or not to the same subject. Initially the face image is segmented into four different regions, which are hierarchically compared according. The hierarchy is defined according to each region's size, arranged from the smallest ones to the biggest ones. At each level of the hierarchy, the similarity measure is evaluated to verify if at that step we can ensure if both faces are from the same subject. With this approach, we can boost the system performance and also reduce its computational time. Experimental results were performed using all images from the FRGC v2 database, and the results show the effectiveness of this approach.