Research On 3D Face Recognition Based On Data Augmentation

Two-dimensional face recognition performs poorly in the face of changes in factors such as light,makeup,posture,and occlusion.In actual applications,the requirements for the collected images are high,and clear frontal images need to be collected.The 3D face data contains depth information that cannot be provided by the 2D face image,so it can effectively deal with the interference caused by changes in light and makeup.The traditional three-dimensional face recognition method is complicated,but the threedimensional face recognition method based on deep learning,because of the lack of a sufficient scale of data sets,currently can not reach the research level of twodimensional face recognition.Based on our research,most of the current research results are based on the existing three-dimensional data to transform expression,makeup and posture,etc.,so as to achieve the effect of enhancing the data set.However,the increased data size is still far from the magnitude of the two-dimensional face dataset,and the network's generalization ability is insufficient.In this paper,a three-dimensional face reconstruction method is used for data enhancement,and the enhanced data set is used to train the network to extract facial features for three-dimensional face recognition.The main contributions of the paper are as follows:(1)Introduce a data enhancement method to increase the identity of 3D face data.This method uses the existing advanced face reconstruction technology and uses an unlimited single 2D face image to reconstruct end-to-end to obtain complete 3D face information,increasing the number of 3D face data identities.(2)Based on the three-dimensional face data,a unique data preprocessing method is designed according to the preprocessing process.We convert the extracted threedimensional face data and map it to RGB space to form a three-channel data form as the input data of the convolutional neural network.(3)Borrow and use the convolutional neural network FR3 DNet adjusted based on VGGNet to train our 3D face recognition network.An end-to-end 3D face recognition system is built through our trained 3D face recognition network.Finally,based on a total of 79,015 two-dimensional face images with 2000 identities,face reconstruction was performed.We train the convolutional neural network on the reconstructed face dataset and verify it on the public dataset.The results show that the recognition effect of this method is better than other methods.We designed and simulated an open world environment,tested the model,and found that our recognition rate also exceeded the results of others.