An Automatic 2D, 2.5D & 3D Score-Based Fusion Face Verification System

A score-based fusion for face verification is presented from FRAV3D Face Database (2D, 2.5D and 3D face images). In the case of 2.5D and 3D data, an automatic correction of pose has been carried out by detecting the nose tip and the eyes. For each kind of image a different feature extraction has been applied (Principal Component Analysis and Support Vector Machine for 2D and 2.5D, and Iterative Closest Point algorithm for 3D). A fusion at score level has been performed two by two, after a minimum-maximum normalization (MM) and a Z-score standardization (ZS). We have found an optimal combination that reduces (or at least does not worsen) the Equal Error Rate of the classifiers applied independently. In the most optimal situation, the improvement of the EER is higher than 80% for the fusion of 2D and 2.5D data, as well as for 2.5D and 3D data.