Multimodal hybrid face recognition based on score level fusion using Relevance Vector Machine

In this paper, we present a novel fusion approach 2D and 3D face recognition based on score level fusion using Relevance Vector Machine. RVM uses a combination of kernel functions on training data for classification and compared to SVM. Several novelties are introduced to make the recognition robust to facial expressions. These novelties include: Automatic invariant feature in the pose and in the facial expressions is calculated from 2D facial image and from the point cloud and to generate a local descriptor. These two descriptors are reduced separately by the Principal Component Analysis to provide two scores which are normalized and then are fused to improve the recognition performance. The proposed approach is tested on a CASIA-3D database, the total number of scans is 4674 among which 38 scans per person are available. The results are found to be promising, and show the potential of our approach, we obtain a 99.4% verification rate at 0.01 FAR.