基于二维纹理重建三维人脸深度图像后的人脸识别

Face Recognition after Reconstructing 3D Face Depth Image Based on Two-Dimensional Texture

利用二维纹理重建三维人脸深度图像,做到重建后的三维人脸深度图像可与人脸库中的三维模型深度图像进行特征提取比对识别。利用自编码网络将二维纹理映射重建为三维深度图,再利用卷积神经网络将二维人脸纹理图像映射的三维深度图与已有人脸库中的三维模型生成的深度图像分别提取特征进行对比识别。三维人脸深度图像对二维人脸纹理图像的异构异质识别率在100人的722组数据上达到75.21%的识别准确率。利用简单的卷积神经网络训练少量数据即可以达到不错的人脸识别准确率,表明基于二维纹理重建三维人脸深度图像后的人脸识别方法的可行性以及在数据增加和网络优化之后准确率存在极大地提升空间。

Reconstruction of 3D face depth images using 2D texture, the reconstructed 3D face depth image can be compared with the 3D model depth image in the face database for feature extraction and recognition. The self-coding network is used to reconstruct the 2D texture mapping into three-dimensional depth map. Then the convolution neural network is used to extract features from the three-dimensional depth map of two-dimensional face texture image mapping and the three-dimensional model generated in the existing face database. The heterogeneous recognition rate of 3D face depth image to 2D face texture image is 75.21% on 722 sets of data of 100 people. Using simple convolution neural network to train a small amount of data can achieve a good face recognition accuracy, which shows the feasibility of face recognition method based on two-dimensional texture reconstruction of 3D face depth image and the great improvement of accuracy after data increase and network optimization.