基于自适应圆边际的深度人脸识别算法

doi:10.3969/j.issn.1000-1093.2021.11.016

摘要/Abstract

摘要： 人脸识别是计算机视觉的一个重要研究方向，其中有效的损失函数在人脸识别中起着至关重要的作用。针对现有损失函数没有考虑边际情况，导致模型收敛有限，且在不均衡样本中泛化能力不强的问题，提出自适应圆边际损失函数方法，对边际自身进行研究。通过对边际进行自适应学习，为不同类别学习独有的边际，产生自适应圆边际。为少量样本学习更大边际，从而对少量样本数据类内压缩更紧凑，使模型泛化能力更强，对5种常见的人脸识别基准Megaface、IJB-C、LFW、LFW BLUFR和YTF进行广泛分析和实验验证。结果表明，该方法在不均衡数据集中对现有方法的精确度整体提高了0.5%，有效提高了模型泛化能力，具有明确的收敛状态。

关键词: 深度人脸识别, 自适应圆边际, 损失函数, 模型泛化能力, 收敛状态

Abstract: Face recognition is an important research direction of computer vision，and the effective loss functions play a vital role in face recognition. In view of the fact that the existing loss function does not consider the marginal situation to result in a limited model convergence and a low generalization ability is for unbalanced samples，AdaCMloss (Adaptive circle Margin Loss) loss function method is proposed for studying the margin itself.Through the self-adaptive learning of the margin，the unique margin can be learnt for different categories，and the self-adaptive circle margin is generated. A more margin is learnt for a small number of samples，so that the intra-class compression of the data of a small number of samples is more compact and the model generalization ability is stronger. The common face recognition benchmarks Megaface, IJB-C, LFW, LFW BLUFR and YTF are extensively analyzed and experimentally varified. The results show that the proposed method is used to improve the convergence accuracy of existing methods by 0.5% in unbalanced data sets and enhance the model generalization ability effectively, and has a clear convergence state.

Key words: deepfacerecognition, adaptivecirclemargin, lossfunction, modelgeneralizationability, convergencestate

中图分类号:

TP391.41

才华，孙俊，朱瑞昆，朱新丽，赵义武. 基于自适应圆边际的深度人脸识别算法[J]. 兵工学报, 2021, 42(11): 2424-2432.

CAI Hua， SUN Jun， ZHU Ruikun， ZHU Xinli， ZHAO Yiwu. Depth Face Recognition Algorithm Based on Adaptive Circle Margin[J]. Acta Armamentarii, 2021, 42(11): 2424-2432.

参考文献

［1］ SCHROFF F，KALENICHENKO D，PHILBIN J. FaceNet: a unified embedding for face recognition and clustering［C］∥Proceedings of 2015 IEEE Conference on Computer Vision and Pattern Recognition. Boston，MA，US:IEEE，2015:815-823.
［2］ ZHANG X，FANG Z Y，WEN Y D，et al. Range loss for deep face recognition with long tailed training data［C］∥Proceedings of 2017 IEEE International Conference on Computer Vision. Venice，Italy:IEEE，2017:5419-5428.
［3］ YIN X，YU X，SOHN K，et al. Feature transfer learning for face recognition with under-represented data［C］∥Proceedings of 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Long Beach，CA，US:IEEE，2019:5704-5713.
［4］ SUN Y F，CHENG C M，ZHANG Y H，et al. Circle loss:a unified perspective of pair similarity optimization［C］∥Proceedings of 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Seattle，WA，US:IEEE，2020: 6398-6407.
［5］ TAIGMAN Y，YANG M，RANZATO M，et al. Deepface: closing the gap to human-level performance in face verification［C］∥Proceedings of 2014 IEEE Conference on Computer Vision and Pattern Recognition. Columbus，OH，US:IEEE，2014:1701-1708.
［6］ SUN Y，WANG X G，TANG X O. Deeply learned face representations are sparse，selective，and robust［C］∥Proceedings of 2015 IEEE Conference on Computer Vision and Pattern Recognition. Boston，MA，US:IEEE，2015:2892-2900.
［7］ ZHENG Y T，PAL D K，SAVVIDES M. Ring loss: convex feature normalization for face recognition［C］∥Proceedings of 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Salt Lake City，UT，US:IEEE，2018: 5089-5097.
［8］ MAZE B，ADAMS J，DUNCAN J A，et al. Iarpa janus benchmark-c:face dataset and protocol［C］∥Proceedings of 2018 International Conference on Biometrics. Gold Coast，QLD，Australia:IEEE，2018:158-165.
［9］张欣彧，尤鸣宇，朱江，等. 基于联合损失函数的小规模数据人脸识别［J］. 北京理工大学学报，2020，40(2):163-168.
ZHANG X Y，YOU M Y，ZHU J，et al. Face recognition of small-scale data based on joint loss function［J］. Transactions of Beijing Institute of Technology，2020，40(2):163-168. (in Chinese)
［10］ GUO Y D，ZHANG L，HU Y X，et al. Ms-celeb-1m: A dataset and benchmark for large-scale face recognition［C］∥Proceedings of European Conference on Computer Vision. Berlin，German: Springer，2016: 87-102.
［11］ KEMELMACHER-SHLIZERMAN I，SEITZ S M，MILLER D，et al. The megaface benchmark: 1 million faces for recognition at scale［C］∥Proceedings of 2016 IEEE Conference on Computer Vision and Pattern Recognition. Las Vegas，NV，US:IEEE，2016:4873-4882.
［12］ PARKHI O M，VEDALDI A，ZISSERMAN A，et al. Deep face recognition［C］∥Proceedings of British Machine Vision Conference. Swansea，UK: the British Machine Vision Association，2015.
［13］ LIU W Y，WEN Y D，YU Z D，et al. Sphereface: deep hypersphere embedding for face recognition［C］∥Proceedings of 2017 IEEE Conference on Computer Vision and Pattern Recognition. Honolulu，HI，US: IEEE，2017:6738-6746.
［14］ SUN Y，CHENG Y H，WANG X G，et al. Deep learning face representation by joint identification verification［J］. Advances in Neural Information Processing Systems，2014，27:1988-1996.
［15］ SUN Y，WANG X G，TANG X O. Deep learning face representation from predicting 10，000 classes［C］∥Proceedings of 2014 IEEE Conference on Computer Vision and Pattern Recognition. Columbus，OH，US:IEEE，2014:1891-1898.
［16］ WEN Y D，ZHANG K P，LI Z F，et al. A discriminative feature learning approach for deep face recognition［C］∥Proceedings of European Conference on Computer Vision. Berlin，German: Springer，2016:499-515.
［17］ WANG F，CHENG J，LIU W Y，et al. Additive margin softmax for face verification［J］ IEEE Signal Processing Letters，2018，25(7): 926-930.
［18］ DENG J K，GUO J，XUE N N，et al. ArcFace:additive angular margin loss for deep face recognition［C］∥Proceedings of 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Long Beach，CA，US:IEEE，2019:4690-4699.
［19］ WOLF L，HASSNER T，MAOZ I. Face recognition in unconstrained videos with matched background similarity［C］∥Proceedings of IEEE Conference on Computer Vision and Pattern Recognition. Colorado Springs，CO，US:IEEE，2011:529-534.
［20］ NG H W，WINKLER S. A data-driven approach to cleaning large face datasets［C］∥Proceedings of 2014 IEEE International Conference on Image Processing. Paris，France:IEEE，2014: 343-347.
［21］ WANG F，XIANG X，CHENG J，et al. NormFace: L₂ hypersphere embedding for face verification［C］∥Proceedings of the 25th ACM International Conference on Multimedia. Mountain View，CA，US:ACM，2017:1041-1049.
［22］ SUN Y F，CHENG C M，ZHANG Y H，et al. Circle loss: a unified perspective of pair similarity optimization［C］∥Proceedings of 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Seattle，WA，US:IEEE，2020:6398-6407.
［23］章东平，陈思瑶，李建超，等. 基于改进型加性余弦间隔损失函数的深度学习人脸识别［J］. 传感技术学报，2019，32(12): 1830-1835.
ZHANG D P，CHEN S Y，LI J C，et al. Deep learning face recognition based on improved additive cosine margin softmax interval loss function［J］. Chinese Journal of Sensors and Actuators，2019，32(12):1830-1835. (in Chinese)
［24］ LIU H，ZHU X Y，LEI Z，et al. AdaptiveFace: adaptive margin and sampling for face recognition［C］∥Proceedings of 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Long Beach，CA，US:IEEE，2019:11947-11956.
［25］ CAO D， HU X Y，HUANG X Y，et al. Domain balancing:face recognition on long-tailed domains［C］ ∥Proceedings of IEEE/CVF International Conference on Computer Vision and Pattern Recognition. Seattle，WA，US:IEEE，2020:5671-5679.
［26］ ZHONG Y Y，DENG W H，WANG M，et al. Unequal-training for deep face recognition with long-tailed noisy data［C］∥Proceedings of the 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Long Beach，CA，US:IEEE，2019:7812-7821.
［27］姬东飞，丁学明. 基于自适应角度损失函数的深度人脸识别算法研究［J］. 计算机应用研究，2020，37(10):3169-3174.
JI D F，DING X M. Research on deep face recognition algorithm based on adaptive angle loss function［J］. Application Research of Computers，2020，37(10):3169-3174. (in Chinese)