中心损失与Softmax损失联合监督下的人脸识别
作者:
中图分类号:

TP391.4

基金项目:

国家自然科学基金资助项目(61402063);重庆市科技人才培养计划资助项目(CSJC2013KJRC-TDJS40012);重庆市高校优秀成果转化资助项目(KJZH14213)。


Joint supervision of center loss and softmax loss for face recognition
Author:
  • 摘要
  • | |
  • 访问统计
  • |
  • 参考文献 [20]
  • |
  • 相似文献
  • | | |
  • 文章评论
    摘要:

    深度学习在人脸识别领域已经取得了巨大的成就,针对当前大多数卷积神经网络采用Softmax损失函数进行特征分类,增加新的类别样本会减小类间距离的增长趋势,影响网络对特征判别的问题,采用了一种基于中心损失与Softmax损失联合监督的人脸识别算法,来提高网络对特征的识别能力。在Softmax基础上,首先,分别对训练集每个类别在特征空间维护一个类中心,训练过程新增加样本时,网络会约束样本的分类中心距离,从而兼顾了类内聚合与类间分离。其次,引入动量概念,在分类中心更新的时候,通过保留之前的更新方向,同时利用当前批次的梯度微调最终的更新方向,该方法可以在一定程度上增加稳定性,提高网络的学习效率。最后,在人脸识别基准库LFW上的测试实验证明:所提的联合监督算法,在较小的网络训练集上,获得了99.31%的人脸识别精度。

    Abstract:

    Nowadays, deep learning has made great achievements in face recognition. Most of the convolutional neural network uses the Softmax loss function to increase the distance between classes. However, adding samples of new classes will reduce the distance between classes and the performance of the network. In order to improve the recognition ability of the network characteristics, a face recognition approach based on joint supervision of center loss and Softmax loss is proposed. On the basis of Softmax, first of all, a class center is maintained in the feature space for each class of the training set. When a new sample is added to the training process, the network will constrain the distance of the classification center of the sample, and thus both intra-class aggregation and inter-class separation are considered. Secondly, the concept of momentum is introduced. When the classification center is updated, by retaining the previous update direction and using the gradient of the current batch to fine-tune the final update direction, the method can increase the stability and improve the learning efficiency of the network. Finally, the test experiments on the face recognition benchmark library LFW (labeled faces in the wild) prove that the proposed joint supervision algorithm achieves 99.31% of face recognition accuracy on a small network training set.

    参考文献
    [1] Parkhi O M, Vedaldi A, Zisserman A, et al. Deep face recognition[C]//British Machine Vision Conference. Swansea, UK:British Machine Vision Association, 2015:41-42.
    [2] Mufti H N, Abidi S, Abidi S R, et al. Predictors of post-operative delirium in cardiac surgery patients; a machine learning approach[J]. Canadian Journal of Cardiology, 2014, 30(10):237-238.
    [3] LeCun L, Bottou Y, Bengio P, et al. Gradient-based learning applied to document recognition[J]. Proceedings of the IEEE, 1998, 86(11):2278-2323.
    [4] Ballester P, Araujo R M. On the performance of GoogLeNet and alexNet applied to sketches[C]//Thirtieth AAAI Conference on Artificial Intelligence. Phoenix, USA:AAAI Press, 2016:1124-1128.
    [5] Day M J, Horzinek M C, Schultz R D, et al. Guidelines for the vaccination of dogs and cats[J]. Journal of Small Animal Practice, 2007, 48(9):528.
    [6] He K, Zhang X, Ren S, et al. Deep residual learning for image recognition[G]//Computer Vision and Pattern Recognition, Las Vegas, USA:IEEE, 2015:770-778.
    [7] Sinno J, Yang Q. A survey on transfer learning[J]. IEEE Transactions on Knowledge & Data Engineering, 2010, 22(10):1345-1359.
    [8] Hadsell R, Chopra S, Lecun Y, et al. Dimensionality reduction by learning an invariant mapping[C]//Computer Vision and Pattern Recognition, Conference on 2006 IEEE Computer Society. New York:IEEE, 2006:1735-1742.
    [9] Chopra S, Hadsell R, Lecun Y, et al. Learning a similarity metric discriminatively, with application to face verification[C]//Computer Vision and Pattern Recognition, 2005. New York:IEEE, 2005:539-546.
    [10] Taigman Y, Yang M. Deepface:closing the gap to human-level performance in face verification[C]//Computer Vision and Pattern Recognition. Columbus, USA:IEEE, 2014:1701-1708.
    [11] Sun Y, Wang X, Tang X. Deep learning face representation from predicting 10,000 classes[C]//IEEE Conference on Computer Vision and Pattern Recognition. Columbus, USA:IEEE Computer Society, 2014:1891-1898.
    [12] Sun Y, Wang X, Tang X, et al. Deep learning face representation by joint identification-verification[C]//NIPS Montred, CANADA:MIT Press, 2014, 27:1988-1996.
    [13] Cheng D, Gong Y, Zhou S, et al. Person reidentification by multi-channel parts-based CNN with improved triplet loss function[C]//Computer Vision and Pattern Recognition. Las Vega S, USA:IEEE, 2016:1335-1344.
    [14] King G, Zeng L. Logistic regression in rare events data[J]. Political Analysis, 2001, 9(2):137-163.
    [15] Wen Y, Zhang K, Li Z, et al. A discriminative feature learning approach for deep face recognition[C]//Computer Vision-ECCV 2016. Amsterdam:Springer International Publishing, 2016:11-26.
    [16] Qian N. On the momentum term in gradient descent learning algorithms[J]. Neural Networks the Official Journal of the International Neural Network Society, 1999, 12(1):145-151.
    [17] Simonyan K, Zisserman A. Very deep convolutional networks for large-scale image recognition[J]. Computer Science, 2014, 6(1):1409-1556.
    [18] Szegedy C, Ioffe S, Vanhoucke V, et al. Inception-v4, Inception-ResNet and the impact of residual connections on learning[C]//AAAI. Palo Alto, California:AAAI, 2017:4278-4284.
    [19] Kazemi V, Sullivan J. One millisecond face alignment with an ensemble of regression trees[C]//Computer Vision and Pattern Recognition. Columbus, USA:IEEE, 2014:1867-1874.
    [20] Klawonn F, H ppner F, May S. An alternative to ROC and AUC analysis of classifiers[C]//International Conference on Advances in Intelligent Data Analysis X. Berlin:Springer, 2011:210-221.
    相似文献
    引证文献
    网友评论
    网友评论
    分享到微博
    发 布
引用本文

余成波,田桐,熊递恩,许琳英.中心损失与Softmax损失联合监督下的人脸识别[J].重庆大学学报,2018,41(5):92-100.

复制
分享
文章指标
  • 点击次数:
  • 下载次数:
  • HTML阅读次数:
  • 引用次数:
历史
  • 收稿日期:2017-11-20
  • 在线发布日期: 2018-05-23
文章二维码