基于生成对抗网络的人脸年龄渐进合成算法

doi:10.12068/j.issn.1005-3026.2024.07.005

摘要/Abstract

摘要：

人脸年龄合成（face age synthesis，FAS）的目标是根据源人脸图像合成指定年龄人脸图像，同时保留人脸的个人特征和身份信息.针对年龄变换时无关特征容易改变和产生伪影鬼影的问题，提出一种基于生成对抗网络的人脸年龄渐进合成算法.采用基于门控循环单元的年龄编辑模块自适应地过滤或加入特征，并使用属性解耦模块在潜在空间进行对抗学习，通过生成器和判别器的对抗策略保证了真实自然的人脸合成，使用年龄分类约束拟合特定年龄分布，为了保证年龄无关属性的保留，还在生成对抗网络中引入了重建学习.在跨年龄名人数据集（cross?age celebrity dataset，CACD）下的实验结果表明，对比其他基于条件生成对抗网络的算法，提出的算法生成的人脸图像伪影失真有所减少，年龄显著性增强，具有较好的年龄准确性和较高的身份一致性.

关键词: 人脸年龄合成, 生成对抗网络, 属性解耦, 潜在空间, 门控循环单元, 重建学习

Abstract:

The goal of face age synthesize （FAS） is to synthesize face images of specified ages based on the source face image， while preserving personal characteristics and identity information of the face. To solve the problem that irrelevant features are easy to change and artifact ghosting occurs when age is changed， a progressive face age synthesis algorithm based on generative adversarial network is proposed. The age editing module based on gate recurrent unit is used to filter or add features adaptively， and attribute decoupling module is used for adversarial learning in the latent space. Through the adversarial strategy of generator and discriminator， the real and natural face synthesis is guaranteed. The age classification constraint is used to fit the specific age distribution. In order to preserve age?independent properties， reconstruction learning is also introduced into generative adversarial network. Experimental results on CACD dataset show that， compared with other algorithms based on conditional generative adversarial network， the proposed algorithm has reduced artifacts and distortions， enhanced age significance， and has better age accuracy and higher identity consistency.

Key words: face age synthesis, generative adversarial network, attribute decoupling, latent space, gate recurrent unit, reconstruction learning

中图分类号:

TP 391

杨晓雨, 王爱侠, 杨钢, 李晶皎. 基于生成对抗网络的人脸年龄渐进合成算法[J]. 东北大学学报（自然科学版）, 2024, 45(7): 944-952.

Xiao-yu YANG, Ai-xia WANG, Gang YANG, Jing-jiao LI. Progressive Face Age Synthesis Algorithm Based on Generative Adversarial Network[J]. Journal of Northeastern University(Natural Science), 2024, 45(7): 944-952.

图/表 11

图1 人脸年龄渐进合成算法框架

Fig.1 Framework of face progressive synthesis algorithm

图2 卷积门控循环单元内部结构图

Fig.2 Diagram of convolution gate recurrent unit internal structure

图3 判别器结构图

Fig.3 Diagram of discriminator structure

表1 CACD数据信息统计

Table 1 CACD data information statistics

年龄组	21~30	31~40	41~50	51~62
训练集	38 620	41 024	37 805	26 534
测试集	2 085	2 111	1 955	1 465

图4 人脸年龄渐进合成结果

Fig.4 Progressive face age synthesis results

图5 与其他人脸年龄渐进合成算法对比

Fig.5 Comparison with other progressive face age synthesis algorithms

表2 年龄分布评估结果

Table 2 Age distribution assessment results

年龄组	21~30	31~40	41~50	51~62
真实人脸	28.43±7.28	35.31±8.83	43.00±11.25	49.88±12.20
合成人脸	29.63±7.94	36.34±9.42	44.43±9.87	48.39±11.95

表3 平均年龄误差比较

Table 3 Comparison of mean age errors

年龄组	21~30	31~40	41~50	51~62
IPCGAN	0.74	1.16	1.55	3.09
SAM-GAN	0.67	1.29	1.46	2.06
PFA-GAN	0.81	1.05	1.51	2.73
本文算法	1.20	1.03	1.43	1.49

表4 客观人脸验证结果

Table 4 Objective face verification results

年龄组	21~30	31~40	41~50	51~62
真实人脸	93.75±1.25	92.58±2.19	89.87±4.17	88.74±4.66
21~30	—	96.20±0.99	93.91±3.12	93.00±3.63
31~40	—	—	—	95.54±1.77
41~50	—	—	—	96.92±0.20

表5 人脸验证率比较 (%)

Table 5 Comparison of face verification rate

年龄组	21~30	31~40	41~50	51~62
IPCGAN	99.99	98.50	97.82	95.94
SAM-GAN	100	99.87	98.39	96.20
PFA-GAN	100	99.99	98.47	96.97
本文算法	100	99.86	98.65	97.12

图6 消融实验结果

Fig.6 Ablation results

参考文献 33

1	Suo J L， Zhu S C， Shan S G，et al.A compositional and dynamic model for face aging［J］.IEEE Transactions on Pattern Analysis and Machine Intelligence，2010，32（3）：385-401.
2	Tazoe Y， Gohara H， Maejima A，et al.Facial aging simulator considering geometry and patch‑tiled texture［M］.New York：ACM，2012.
3	Tsai M H， Liao Y K， Lin I C.Human face aging with guided prediction and detail synthesis［J］.Multimedia Tools and Applications，2014，72（1）：801-824.
4	Todd J T， Mark L S， Shaw R E，et al.The perception of human growth［J］.Scientific American，1980，242（2）：132-145.
5	Suo J L， Chen X L， Shan S G，et al.A concatenational graph evolution aging model［J］.IEEE Transactions on Pattern Analysis and Machine Intelligence，2012，34（11）：2083-2096.
6	Ramanathan N， Chellappa R.Modeling age progression in young faces［C］//2006 IEEE Computer Society Conference on Computer Vision and Pattern Recognition.New York，2006：387-394.
7	Kemelmacher‑Shlizerman I， Suwajanakorn S， Seitz S M.Illumination‑aware age progression［C］//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition.Columnus，2014：3334-3341.
8	Tiddeman B， Burt M， Perrett D.Prototyping and transforming facial textures for perception research［J］.IEEE Computer Graphics and Applications，2001，21（5）：42-50.
9	Wang W， Cui Z， Yan Y，et al.Recurrent face aging［C］//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition.Las Vegas，2016：2378–2386.
10	Duong C N， Luu K， Quach K G，et al.Longitudinal face modeling via temporal deep restricted boltzmann machines［C］//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition.Las Vegas，2016：5772-5780.
11	Duong C N， Quac K G， Luu K，et al.Temporal non‑volume preserving approach to facial age‑progression and age‑invariant face recognition［C］//Proceedings of the IEEE International Conference on Computer Vision.Venice，2017：3735-3743.
12	Goodfellow I， Pouget‑Abadie J， Mirza M，et al.Generative adversarial nets［J］.Advances in Neural Information Processing Systems，2014，27（2）：2672-2680.
13	Zhu J Y， Park T， Isola P，et al.Unpaired image‑to‑image translation using cycle consistent adversarial networks［C］//Proceedings of the IEEE International Conference on Computer Vision.Venice，2017：2223-2232.
14	Isola P， Zhu J Y， Zhou T H，et al.Image‑to‑image translation with conditional adversarial networks［C］//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition.Honolulu，2017：1125-1134.
15	Choi Y， Uh Y， Yoo J，et al.Stargan v2：diverse image synthesis for multiple domains［C］//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition.Seattle，2020：8188-8197.
16	Park T， Liu M Y， Wang T C，et al.GauGAN：semantic image synthesis with spatially adaptive normalization［M］.New York：ACM，2019.
17	Karras T， Aittala M， Laine S，et al.Alias‑free generative adversarial networks［J］.Advances in Neural Information Processing Systems，2021，34（1）：852-863.
18	Mirza M， Osindero S.Conditional generative adversarial nets ［EB/OL］.（2014-10-06）［2023-03-04］..
19	Antipov G， Baccouche M， Dugelay J L.Face aging with conditional generative adversarial networks［C］//2017 IEEE International Conference on Image Processing.Beijing，2017：2089-2093.
20	Zhang Z F， Song Y， Qi H R.Age progression/regression by conditional adversarial autoencoder［C］//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition.Honolulu，2017：5810-5818.
21	Wang Z W， Tang X， Luo W X，et al.Face aging with identity‑preserved conditional generative adversarial networks［C］//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition.Salt Lake City，2018：7939-7947.
22	Yang H Y， Huang D， Wang Y H，et al.Learning face age progression：a pyramid architecture of gans［C］//IEEE Conference on Computer Vision and Pattern Recognition.Salt Lake City，2018：31-39.
23	Yang H Y， Huang D， Wang Y H，et al.Learning continuous face age progression：a pyramid of gans［J］.IEEE Transactions on Pattern Analysis and Machine Intelligence，2021，43（2）：499-515.
24	Liu Y F， Li Q， Sun Z N.Attribute‑aware face aging with wavelet‑based generative adversarial networks［C］//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition.Long Beach，2019：11877-11886.
25	Shao J， Bui T D.Wavelet‑based multi‑level generative adversarial networks for face aging［C］//The 32nd British Machine Vision Conference.Virtual：British Machine Vision Association，2021：1388-1399.
26	Li P P， Hu Y B， Li Q，et al.Global and local consistent age generative adversarial networks［C］//International Conference on Pattern Recognition.Beijing，2018：1073-1078.
27	Huang Z Z， Chen S Z， Zhang J P，et al.PFA‑GAN：progressive face aging with generative adversarial network［J］.IEEE Transactions on Information Forensics and Security，2020，15（16）：2031-2045.
28	Li Q， Liu Y F， Sun Z N.Age progression and regression with spatial attention modules［C］//Proceedings of the AAAI Conference on Artificial Intelligence.New York：AAAI，2020：11378-11385.
29	Huang Z Z， Zhang J P， Shan H M.When age‑invariant face recognition meets face age synthesis：a multi‑task learning framework［C］//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition.Virtual，2021：7282-7291.
30	Lample G， Zeghidour N， Usunier N，et al.Fader networks：manipulating images by sliding attributes［C］//Annual Conference on Neural Information Processing Systems.Long Beach：MIT Press，2017：5963-5972.
31	Arjovsky M， Chintala S， Bottou L.Wasserstein GAN ［EB/OL］.（2017-01-26）［2023-03-04］..
32	Gulrajani I， Ahmed F， Arjovsky M，et al.Improved training of Wasserstein gans［C］//Annual Conference on Neural Information Processing Systems.Long Beach：MIT Press，2017：5769-5779.
33	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.

[1]	刘纪红, 张律恒, 杨海旭. 一种细胞荧光显微图像饱和伪影修复算法[J]. 东北大学学报（自然科学版）, 2024, 45(7): 921-927.
[2]	孙颖，李泽，张雪英. 基于约束式双通道模型的语音情感识别[J]. 东北大学学报（自然科学版）, 2023, 44(11): 1537-1542.
[3]	张雪峰，许华文，杨棉子美. 一种基于条件生成对抗网络的高感知图像压缩方法[J]. 东北大学学报（自然科学版）, 2022, 43(6): 783-791.
[4]	李大鹏，赵琪珲，邢铁军，赵大哲. 基于分层注意力循环神经网络的司法案件刑期预测[J]. 东北大学学报（自然科学版）, 2022, 43(3): 344-349.
[5]	罗家健，冯宝，陈相猛，顾正晖. 基于生成对抗网络的肺结节良恶性诊断算法[J]. 东北大学学报（自然科学版）, 2022, 43(1): 24-32.
[6]	李晨，张家伟，张昊，汪茜. 基于生成对抗网络的低分化宫颈癌病理图像分类[J]. 东北大学学报:自然科学版, 2020, 41(7): 1054-1061.
[7]	孟琭，钟健平，李楠. 基于GAN的医学图像仿真数据集生成算法[J]. 东北大学学报:自然科学版, 2020, 41(3): 332-336.
[8]	魏颖, 徐楚翘, 刁兆富, 李伯群. 基于生成对抗网络的多目标行人跟踪算法[J]. 东北大学学报（自然科学版）, 2020, 41(12): 1673-1680.
[9]	徐林，郑晓彤，付博，田歌. 基于改进GAN算法的电机轴承故障诊断方法[J]. 东北大学学报:自然科学版, 2019, 40(12): 1679-1684.
[10]	徐久强，洪丽萍，朱宏博，赵海. 一种用于肺结节恶性度分类的生成对抗网络[J]. 东北大学学报:自然科学版, 2018, 39(11): 1556-1561.