角度余量损失和中心损失联合的深度人脸识别  

Deep face recognition combined with angular margin loss and center loss

作　　者：李振东[1,2] 钟勇 陈蔓[1,2] 王理顺 LI Zhendong;ZHONG Yong;CHEN Man;WANG Lishun(Chengdu Institute of Computer Application,Chinese Academy of Sciences,Chengdu Sichuan 610041,China;University of Chinese Academy of Sciences,Beijing 100049,China)

机构地区：[1]中国科学院成都计算机应用研究所,成都610041 [2]中国科学院大学,北京100049

出　　处：《计算机应用》2019年第S02期55-58,共4页journal of Computer Applications

基　　金：四川省科技厅项目(2018GZ0231)

摘　　要：针对当前人脸识别任务和人脸验证任务所面临的人脸识别准确率低、验证有误等安全隐患问题,为进一步增强深度特征的判别能力,提出了一种联合监督信号。首先,将角度余量损失和中心损失联合的损失函数用于人脸识别和人脸验证任务,实现最小化类内的变化以及最大化类之间的距离,这种深度特征之间的最小化和最大化可以达到人脸面部识别任务的理想选择。所提方法结合深度卷积神经网络在VGGFace2人脸数据集上对网络模型进行训练并在明星人脸数据集(CFS)和LFW人脸数据集上进一步微调网络模型进行人脸识别验证。实验结果表明,所提出的损失函数与传统的Softmax损失函数相比能够正确识别不同的人脸图像,而且进一步实现了不同类别之间的面部特征的最大可分离性。Face recognition tasks and face verification tasks are faced with the problem of low security accuracy and false verification, and a joint surveillance signal was proposed to further enhance the discriminative ability of depth features. Firstly, the loss function of combining angular margin loss and center loss was used for face recognition and face verification tasks to minimize the variation within the class and maximize the distance between the classes. The minimum and maximization between such depth features can be ideal for facial recognition tasks. The proposed method combined the deep convolutional neural network to train the network model on the VGGFace2 face dataset and further fine-tune the network model on the Celebrities Face Set(CFS) and Labeled Faces in the Wild(LFW) face datasets for face recognition verification. The experimental results show that, the proposed loss function can correctly identify different face images compared with the traditional Softmax loss function, and further realize the maximum separability of face features between different categories.

关 键 词：人脸识别 深度学习 卷积神经网络 角度余量损失 中心损失 

分 类 号：TP399[自动化与计算机技术—计算机应用技术]