Multi-modal3D Face Recognition Based On Scoring Mechanism And Features Fusion

3D face recognition has become more attractive in pattern recognition field in re-cent years.3D face model can not only provide the texture information of the face, but also provide the geometric information, which demonstrate the human face more compre-hensively. We expect to utilize the different information implicated in the3D data for multi-modal fusion to enhance the effect of face recognition. We utilize the automatically aligned3D face model to generate the corresponding depth and intensity images which are regarded as two different modal data in this thesis. We use the different modal data to extract different features and fuse them in the score level. The global features of face reflect its contours, whereas local features reflect the details. They play different roles in the recognition and have their own advantages. In order to make full use of different-data and characteristics reflected in the face we use; the global feature of range1image and the local feature of intensity image for fusion. Gabor filter has good spatial location and orientation selectivity, which demonstrates good local characteristics. We select the Gabor feature as the local feature and extract the global feature through PCA. Different matching similarity measures have different meaning. In order to eliminate the difference we propose a unified measure of matching-scoring mechanism which is similar to the scoring mechanism utilized in the game. We sort the galleries according their similarities to the unidentified data and assign them the scores according to their ranks. In addi-tion, different features play different roles in the recognition, so it have different weight for integration. Finally, the recognition result is determined according to the the integral weighted sum of the score got in different feature data. The algorithm based on scoring mechanism and feature fusion performs well in the CASIA3D face database.