基于LBP和数据扩充的CNN人脸识别研究

针对卷积神经网络在人脸识别存在的数据集比较少,容易发生过拟合的问题,提出对人脸进行局部二值模式处理,提升图像特征,再引入深度卷积生成对抗网络对局部二值化的人脸进行生成,有效扩充数据集,提升卷积神经网络的泛化能力。该人脸识别卷积神经网络模型包括3层卷积层,3层池化层,1个全连接层,1个Softmax分类回归层。仿真实验中,选取ORL人脸数据库中40人每人10张的人脸图像按8∶1∶1比例设置为训练集、验证集和测试集,并选取Yale人脸数据库中15人每人11张的人脸图像按9∶1∶1的比例设置训练集、验证集和测试集,通过LBP算法提取人脸纹理特征对其进行生成,分别扩充数据集至990张和2200张。结果表明,该算法的人脸识别率不仅高于未扩充数据PCA和LBP等传统人脸识别方法的识别率,而且也将卷积神经网络的识别率提升了约2%,有效提高了泛化能力。

Research on Face Recognition of CNN Based on LBP and Data Expansion

In view of the problem that convolutional neural network has few data sets in face recognition and is prone to over-fitting,the local binary mode processing is carried out to enhance the image features,and the deep convolution generation is introduced to generate the anti-network to generate the local binarized face,which effectively expands the data set and improves the generalization of the convolutional neural network.This convolutional neural network model of face recognition consists of 3-layer convolutional layer,3-layer pooling layer,a fully connected layer and a Softmax classification regression layer.In the simulation experiment,the 10 face images of each of 40 people in the ORL face database are selected as the training set,verification set and test set according to the ratio of 8∶1∶1,and 11 face images of each of 15 people in the Yale face database are selected as the training set,verification set and test set according to the ratio of 9∶1∶1.The face texture features are extracted by LBP algorithm to generate them,and the data set is expanded to 990 sheets and 2200 sheets respectively.The results show that the face recognition rate of the proposed algorithm is not only higher than that of the traditional face recognition methods such as unexpanded data PCA and LBP,but also the recognition rate of the convolutional neural network is increased by about 2%,which effectively improves its generalization.