Research On 3D Face Recognition Technology Based On Multi - Scale Information Fusion

Face recognition has become promising in the area of bio-metric authentication because of its advantages such as noncontact, uniqueness and initiative.2D face recognition technologies have been developed maturely over the last decades, and widely applied in our life. Because 2D image is only the projection of 3D image in 2D space, it is difficult to overcome the influence of illumination, pose and facial expression. There are more face information in 3D image, for example, surface distance information, surface curve information, spatial structure information. So the 3D face recognition has become a hot-topic.In this thesis, a 3D face recognition algorithm fusing Athropometric and Curvelet information. The main work of this thesis includes:1. Made a summary for existing well-known 3D face recognition algorithms. They can be classified into as the follow:model based methods, feature based methods and multimodal fusion methods. And list the existing common 3D face database.2. Proposed applying Anthropometric feature to location the facial feature region. Located the facial key point in the depth image, and used the anthropometric feature and position of the facial key points to locate the facial feature region.3. Extracted feature based on Curvelet transform. Curvelet transform is performed in the facial feature region, then obtain the Curvelet coefficient in muti-scale and muti directions. The feature vector of the feature region is obtained by l1 norm processing used in Curvelet coefficient matrix.4. Obtained the experiment result by Experimenting in the Texas 3DFRD database. The final recognition results are obtained by fusing the recognition results of the four feature region. Compared with the algorithm what only use Anthropometric feature or Curvelet feature, the experiment result show that the algorithm in this thesis can work well.