基于面部表情GEM和稀疏立方矩阵的三维人脸识别方法

针对姿态和表情变化对3D人脸识别影响较大的问题,提出一种基于面部表情通用弹性模型(generic elastic models,GEM)和稀疏立方矩阵的3D人脸识别方法.利用面部表情通用弹性模型构造3D人脸数据库,3D重建模型为所有人脸姿态创建稀疏立方矩阵(sparse cubic matrix,SCM),并利用自动头部姿态估计法获得人脸图像中三元组角度的初始估计值;为每个子集估计的三元组角度选择SCM的阵列;通过稀疏表示从SCM中选择阵列与探针图像.在FERET,CMU PIE和LFW数据库上的实验验证了提出方法的有效性.与几种优秀3D人脸识别方法相比,提出的方法识别率更高,当姿态变化角度较大时尤为明显.此外,对于480×640图像,LFW数据库上,预处理、人脸检测和分类的总平均处理时间仅为89.4 ms.

3D face recognition method based on facial expression generic elastic model and sparse cubic matrix

Concerning the great effect of gesture changing and facial expression changing on 3D face recognition, a 3D face recognition method based on facial expression generic elastic model and sparse cubic matrix is proposed. Firstly, 3D face database is constructed by facial expression generic elastic model, and sparse cubic matrix (SCM) for all face poses is created by 3D reconstruction model, then the initial estimation of triplet angles in face images is obtained by using an automatic head pose estimation method; Then, the triplet angle of each subset is estimated, and SCM array is chosen; Finally, the array and probe image are selected by sparse representation from SCM. The effectiveness of proposed method is verified by experiments on FERET, CMU PIE and LFW database. Compared with several outstanding 3D face recognition methods, the proposed method achieves higher recognition rate, especially in dealing with large angle posture change. In addition, the average processing time of pretreatment, face detection and classification is only 89.4 ms for images size of 480×640 on LFW database.