[1]叶继华,郭祺玥,江爱文,等.基于特征子空间直和的跨年龄人脸识别方法[J].郑州大学学报(工学版),2021,42(05):7-12.[doi:10.13705/j.issn.1671-6833.2021.05.002]
　Ye Jihua,Guo Qiyi,Jiang Aiwen,et al.Cross-age Face Recognition Method Based on Feature Subspace Direct Sum[J].Journal of Zhengzhou University (Engineering Science),2021,42(05):7-12.[doi:10.13705/j.issn.1671-6833.2021.05.002]
点击复制

基于特征子空间直和的跨年龄人脸识别方法()
分享到：

《郑州大学学报(工学版)》[ISSN:1671-6833/CN:41-1339/T]

卷:
42卷
期数:
2021年05期
页码:
7-12
栏目:
出版日期:
2021-09-10

文章信息/Info

Title:
Cross-age Face Recognition Method Based on Feature Subspace Direct Sum
作者:
叶继华; 郭祺玥; 江爱文; 黎欣
江西师范大学计算机信息工程学院;
Author(s):
Ye Jihua; Guo Qiyi; Jiang Aiwen; Li Xin;
School of Computer Information Engineering, Jiangxi Normal University;
关键词:
Keywords:
face recognition;  cross-age;  multi-task;  subspace direct sum;  feature subspace
DOI:
10.13705/j.issn.1671-6833.2021.05.002
文献标志码:
A
摘要:
近年来，学术界提出了许多判别性方法用于解决跨年龄人脸识别任务，并取得了可观的成果。然而，这些方法都在一定程度上忽视了年龄因子与身份因子之间的相关性。因此，本文在同时进行人脸身份识别和年龄分类这两个任务的多任务卷积神经网络的基础上引入直和模块，提出了一种基于特征子空间直和的多任务卷积神经网络 (Feature Subspace with Direct Sum CNN, FSDS-CNN)。该网络利用两个并行子网分别从卷积单元共享的深度特征中提取出身份相关特征和年龄相关特征，并对这两个相关特征所对应的特征子空间施加直和约束，使得身份相关特征与年龄相关特征尽可能无关。通过多损失的联合监督学习，该网络可以获得随年龄变化鲁棒的年龄无关人脸身份特征。本文在三个公开的基准老化数据集上进行了实验并与近几年的10种具有代表性的方法做了对比，在Morph Album 2数据集中，本文方法在Rank-1识别率（Rank-1 Identification Rate）上结果为98.41%，取得了次优值；在CACD-VS数据集中，本文方法在精确度（Accuracy）上结果为99.2%，取得了次优值，在AUC（Area Under Curve）上结果为99.7%，取得了最优值，比性能第2的模型提高了0.1%；在Cross-Age LFW数据集中，本文方法在等错误率（Equal Error Rate, EER）上结果为10.1%，在错误匹配率为0.1时的错误非匹配率（false non-match rate when false match rate is 10%, FNMR@FMR=0.1）上结果为10.2%，均取得了最优值，比性能第2的模型分别下降了4.7%和11.6%。同时本文在三个数据集上的实验均进行了消融对比实验以验证直和模块的有效性，实验结果证明了直和模块的有效性和优越性。本文提出的FSDS-CNN模型利用直和模块有效降低了身份特征与年龄特征的相关性，能够有效提升跨年龄人脸识别的性能。
Abstract:
To solve cross-age face recognition tasks, this paper introduces the direct sum module on the basis of the multi-task convolutional neural network that simultaneously performs two tasks of face recognition and age classification, and proposes the feature subspace with direct sum multi-task convolutional neural network (FSDS-CNN). The network uses two parallel subnets to extract the identity-related feature and age-related feature from the deep feature, then the direct sum constraint is applied to the feature subspaces corresponding to these two related features, so that the correlation between identity-related feature and age-related feature is decreased as much as possible. Through the joint supervised learning of multiple loss functions, the network can obtain age-invariant face identity feature that is robust with age. Cross-age face recognition and verification experiments is conducted on three datasets (Morph Album 2, CACD-VS and Cross-Age LFW). In the CACD-VS dataset, the proposed method achieves the optimal result of 99.7% on the evaluation metric of AUC; in the Cross-Age LFW dataset, the method respectively achieves the optimal results of 10.1% and 10.2% on the evaluation metric of EER and FNMR when FMR is 0.1. At the same time, the ablation comparison experiments are conducted on the three datasets to verify the effectiveness of the direct sum module. The results show that the correlation between identity features and age features is effectively reduced by the direct sum module in FSDS-CNN, and then effectively improves the performance of cross-age face recognition.

参考文献/References:

[1] SAWANT M M,BHURCHANDI K M.Age invariant face recognition:a survey on facial aging databases,techniques and effect of aging[J].Artificial intelligence review,2019,52(2):981-1008.

[2] 董锁芹.基于生成对抗网络的跨年龄人脸识别技术研究[D].长春:长春理工大学,2019.
[3] WEN Y D,LI Z F,QIAO Y.Latent factor guided convolutional neural networks for age-invariant face recognition[C]//2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR).Piscataway:IEEE,2016:4893-4901.
[4] 苏士美,王燕,王明霞.基于加权小波分解的人脸识别算法研究[J].郑州大学学报(工学版),2014,35(1):5-9.
[5] WANG Y T,GONG D H,ZHOU Z,et al.Orthogonal deep features decomposition for age-invariant face re-cognition[C]//Computer Vision-ECCV 2018. Munich, Germany: ECCV, 2018:11219.
[6] LI H X,HU H F,YIP C.Age-related factor guided joint task modeling convolutional neural network for cross-age face recognition[J].IEEE transactions on information forensics and security,2018,13(9):2383-2392.
[7] WANG H,GONG D H,LI Z F,et al.Decorrelated adversarial learning for age-invariant face recognition[C]//2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Piscataway:IEEE,2019:3522-3531.
[8] 叶继华,万叶晶,刘长红,等.基于多子空间直和特征融合的人脸识别算法[J].数据采集与处理,2016,31(1):102-107.
[9] 孙宗明,李振国,梅门昌.维数公式与子空间直和的等价条件[J].长沙大学学报,1998(2):47-50.
[10] ZHANG K P,ZHANG Z P,LI Z F,et al.Joint face detection and alignment using multitask cascaded convolutional networks[J].IEEE signal processing letters,2016,23(10):1499-1503.
[11] YI D,LEI Z,LIAO S C,et al.Learning face representation from scratch[EB/OL]. (2014-11-28)[2020-08-31]. https://arxiv.org/pdf/1411.7923.pdf.
[12] RICANEK K,TESAFAYE T.MORPH:a longitudinal image database of normal adult age-progression[C]//7th International Conference on Automatic Face and Gesture Recognition (FGR06).Piscataway:IEEE,2006:341-345.
[13] CHEN B C,CHEN C S,HSU W H.Face recognition and retrieval using cross-age reference coding with cross-age celebrity dataset[J].IEEE transactions on multimedia,2015,17(6):804-815.
[14] ZHENG T Y,DENG W H,HU J N.Cross-Age LFW:a database for studying cross-age face recognition in unconstrained environments[EB/OL]. (2017-08-28)[2020-08-31].https://arxiv.org/pdf/1708.08197.pdf.
[15] CHEN B C,CHEN C S,HSU W H.Cross-age reference coding for age-invariant face recognition and retrieval[C]//Computer Vision-ECCV 2014. Cham: Springer, 2014:768-783.
[16] GONG D H,LI Z F,TAO D C,et al.A maximum entropy feature descriptor for age invariant face recognition[C]//2015 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). Piscataway:IEEE,2015:5289-5297.
[17] LI Z F,GONG D H,LI X L,et al.Aging face recognition:a hierarchical learning model based on local patterns selection[J].IEEE transactions on image processing,2016,25(5):2146-2154.
[18] ZHENG T Y,DENG W H,HU J N.Age estimation guided convolutional neural network for age-invariant face recognition[C]//2017 IEEE Conference on Computer Vision and Pattern Recognition Workshops (CVPRW).Piscataway:IEEE,2017:503-511.
[19] LING C X, HUANG J, ZHANG H. AUC: a better measure than accuracy in comparing learning algorithms[C]// Advances in Artificial Intelligence, 16th Conference of the Canadian Society for Computational Studies of Intelligence, AI 2003. Cham: Springer,2003:1-25.
[20] CHEN D, CAO X D,WEN F,et al.Blessing of dimensionality:high-dimensional feature and its efficient compression for face verification[C]//2013 IEEE Conference on Computer Vision and Pattern Recognition.Piscataway:IEEE,2013:3025-3032.
[21] PARKHI O M, VEDALDI A, ZISSERMAN A. Deep face recognition[C]//British Machine Vision Conference 2015. Swansea, UK: BMVA, 2015:1-12.
[22] CHEN B H, DENG W H, DU J P. Noisy softmax: improving the generalization ability of DCNN via postponing the early softmax saturation[C]//IEEE Conference on Computer Vision and Pattern Recognition 2017. Piscataway: IEEE,2017: 4021-4030.

更新日期/Last Update: 2021-10-11