2D And 3D Face Image Fusion

In economic, security, social security, crime, military other fields, face recognition has huge potential value. Face Recognition has called the most promising authentication methods in the 21th century because of without user excessive participation, contactless data acquisition manner, no damage to user, easy concealment advantages. Face Recognition could be divided into two and three dimensions face recognition accordanced information types used. If we look on face as plane images viewed it is dimensional identification problem, if we use stereo image to show face it is 3D recognition.At present, the 2D face recognition technology is relatively mature, but most methods have focused on single 2D image recognition. And it is inevitably subjected to the environment (lighting, backgrounds, perspectives, etc.) and the face itself (posture, facial expression, occlusion, etc.) in face identifying. The recognition accuracy is limit. 3D face recognition can be overcome and mitigate these factors, therefore 3D face recognition technology has more and more attention. The multi-modal face identification which compromised 2D and 3D information is expected to achieve better recognition results and it becomes an important recognition of the direction of development.In this paper, the theory of 2D and 3D face recognition and methods have been fused, the focus is the classification of 3D face, that is 3D face gender recognition, and 2D and 3D face image fusion method. The face database used in this article is CASIA 3D face database. First, we preprocess the images. We adjust 3D image rotation which combined 2D images, and we get normalization of face gray image though the preprocessing of 2D images. Then we get 3D depth image though the preprocessing of 3D images. Then, we use the depth image of face to do gender recognition, we mainly used support vector machine (SVM) method based on depth value, the SVM method based on PCA feature and the AdaBoost method based on PCA feature. Comparison with experimental performance of these three methods can be the following conclusions: gender information is distributed in a compact face of a smaller sub-sample space, can sex before PCA for feature extraction characteristics as gender. This will compress the sample dimensionality reduction, and improve the classification speed, and accuracy does not suffer much loss. In the 2D and 3D face fusion, we extract the PCA features from 2D and 3D face images. Then we use four methods to fuse them. Finally, we use the nearest neighbor method to do face recognition. From the experimental results, we can see that 2D and 3D integration method can greatly improve recognition performance.