计算机科学

计算机科学 ›› 2026, Vol. 53 ›› Issue (8): 209-218.doi: 10.11896/jsjkx.250500082

• 计算机图形学 & 多媒体 • 上一篇    下一篇

动态层级特征生成蒸馏的轻量化低分辨率人脸识别

钟锐, 严鸿炜, 刘嘉伟   

  1. 赣南师范大学数学与计算机科学学院 江西 赣州 341000
  • 收稿日期:2025-05-20 修回日期:2025-11-20 出版日期:2026-08-15 发布日期:2026-08-17
  • 通讯作者: 钟锐(zhongrui@gnnu.edu.cn)
  • 基金资助:
    国家自然科学基金(62266003);江西省自然科学基金(20232BAB202056);江西省教育厅科技项目(GJJ211401);江西省基础教育研究课题(SZUGSXX2024-1068)

Lightweight Low-resolution Face Recognition via Hierarchical Dynamic Feature Generation Distillation

ZHONG Rui, YAN Hongwei, LIU Jiawei   

  1. School of Mathematics and Computer Science, Gannan Normal University, Ganzhou, Jiangxi 341000, China
  • Received:2025-05-20 Revised:2025-11-20 Published:2026-08-15 Online:2026-08-17
  • About author:ZHONG Rui,born in 1984,Ph.D,asso-ciate professor,is a member of CCF(No.R4389M).His main research interests include computer vision and pattern recognition.
  • Supported by:
    National Natural Science Foundation of China(62266003), Jiangxi Provincial Natural Science Foundation(20232BAB202056),Science and Technology Project of Jiangxi Provincial Department of Education(GJJ211401) and Jiangxi Basic Education Research Project(SZUGSXX2024-1068).

PDF (PC)

100

摘要： 针对低分辨率人脸识别场景中模型应具备准确、轻量与高效的性能需求,提出了一种动态层级特征生成蒸馏框架(Hierarchical Dynamic Feature Generation Distillation,HDFGD)。该框架基于特征冗余性理论与知识蒸馏技术,构建动态层级特征生成模块以增强轻量化学生网络的特征表达能力, 通过在动态层级特征生成模块中引入带通道拓展的深度卷积来扩展特征多样性,并借助通道注意力机制聚焦关键语义信息,结合自适应压缩机制降低计算复杂度,有效缓解跨分辨率特征间的语义鸿沟。同时,使用ArcFace损失函数来优化特征空间的角度间隔约束, 从而强化类内聚合与类间分离特性,提高低分辨率人脸识别准确性。实验结果表明,HDFGD框架在CIFAR-100分类任务中相比主流方法实现了最高2.1%的准确率提升,同时在低分辨率人脸识别领域的两项基准测试中分别取得95.87%与97.02%的识别精度,明显优于现有轻量化人脸识别算法。消融实验证实,动态特征生成与分层对齐策略结合ArcFace判别性监督机制,能够同步实现模型轻量化、跨分辨率特征鲁棒性与识别精度强化的三重突破,为资源受限场景下的实时人脸识别任务提供高效技术方案。

关键词: 低分辨率人脸识别, 知识蒸馏, 动态特征生成, 轻量化, 分层

Abstract: For low-resolution face recognition scenarios demanding high accuracy,efficiency,and lightweight design,this study proposes a HDFGD(Hierarchical Dynamic Feature Generation Distillation) framework.Based on feature redundancy theory and knowledge distillation,the framework constructs a dynamic hierarchical feature generation module to enhance the feature representation capabilities of lightweight student networks.The module expands feature diversity through channel-wise convolutions,employs channel attention mechanisms to focus on critical semantic information,and integrates adaptive compression mechanisms to reduce computational complexity,effectively mitigating semantic gaps across resolution-specific features.Simultaneously,the ArcFace loss function is incorporated to optimize angular-margin constraints in the feature space,thereby strengthening intra-class compactness and inter-class separability,and ultimately improving the accuracy of low-resolution face recognition.Experimental results demonstrate that the HDFGD framework achieves a maximum 2.1% improvement in accuracy over mainstream methods on the CIFAR-100 classification task.It also attains recognition accuracies of 95.87% and 97.02% respectively on two benchmark low-resolution face re-cognition datasets,significantly outperforming existing lightweight face recognition algorithms.Ablation studies confirm that the combination of dynamic feature generation with hierarchical alignment strategies and the ArcFace discriminative supervision mechanism simultaneously achieves triple breakthroughs:model lightweighting,cross-resolution feature robustness,and recognition accuracy enhancement.This framework provides an efficient technical solution for real-time face recognition tasks in resource-constrained scenarios.

Key words: Low-resolution face recognition, Knowledge distillation, Dynamic feature generation, Lightweight, Hierarchical

中图分类号: 

  • TP391
[1] HE K,ZHANG X,REN S,et al.Deep residual learning forimage recognition[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition.IEEE,2016:770-778.
[2] SCHROFF F,KALENICHENKO D,PHILBIN J.FaceNet:AUnified Embedding for Face Recognition and Clustering[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition.IEEE,2015:815-823.
[3] GROTHER P,NGAN M,HANAOKA Y,et al.Face Recognition Vendor Test(FRVT) Part 3:Demographic Effects[R].Gaithersburg:National Institute of Standards and Technology,2021.
[4] ZHANG X,KU S P.Facial super-resolution reconstructionmethod based on generative adversarial networks[J].Journal of Jilin University(Engineering and Technology Edition),2025,55(1):333-338.
[5] BISWAS S,BOWYER K W,FLYNN P J.Multidimensionalscaling for matching low-resolution face images[J].IEEE Transactions on Pattern Analysis and Machine Intelligence,2012,34(10):2019-2030.
[6] DONG C,LOY C C,HE K,et al.Image Super-Resolution Using Deep Convolutional Networks[C]//Proceedings of the European Conference on Computer Vision.Springer,2014:184-199.
[7] WANG X,YU K,WU S,et al.ESRGAN:Enhanced Super-Re-solution Generative Adversarial Networks[C]//Proceedings of the European Conference on Computer Vision.Springer,2018:63-79.
[8] YU X,FERNANDO B,GHANEM B,et al.Face super-resolution guided by facial component heatmaps[C]//Proceedings of the European Conference on Computer Vision.Springer,2018:217-233.
[9] REN C X,DAI D Q,YAN H.Coupled kernel embedding for low-resolution face image recognition[J].IEEE Transactions on Image Processing,2012,21(8):3770-3783.
[10] HINTON G,VINYALS O,DEAN J.Distilling the Knowledgein a Neural Network[J].Computer Science,2015,14(7):38-39.
[11] KULLBACK S,LEIBLER R A.On information and sufficiency[J].The Annals of Mathematical Statistics,1951,22(1):79-86.
[12] ROMERO A,BALLAS N,KAHOU S E,et al.FitNets:Hints for Thin Deep Nets[C]//Proceedings of the International Conference on Learning Representations.2015.
[13] GE S,ZHAO S,LI C,et al.Low-resolution face recognition in the wild via selective knowledge distillation[J].IEEE Transactions on Image Processing,2019,28(4):2051-2062.
[14] HOWARD A,SANDLER M,CHU G,et al.Searching for MobileNetV3[C]//Proceedings of the IEEE International Confe-rence on Computer Vision.IEEE,2019:1314-1324.
[15] GUO J,CHEN M,HU Y,et al.Reducing the Teacher-Student Gap via Spherical Knowledge Distillation[J].arXiv:2010.07485,2020
[16] CHI Z,ZHENG T,LI H,et al.NormKD:Normalized Logits for Knowledge Distillation[J].arXiv:2308.00520,2023.
[17] HAN K,WANG Y,TIAN Q,et al.GhostNet:More Featuresfrom Cheap Operations[C]//Proceedings of the IEEE Confe-rence on Computer Vision and Pattern Recognition.IEEE,2020:1580-1589.
[18] DENG J,GUO J,XUE N,et al.ArcFace:Additive Angular Margin Loss for Deep Face Recognition[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition.IEEE,2019:4690-4699.
[19] SINHA D,EL-SHARKAWY M.Thin MobileNet:An Enhanced MobileNet Architecture[C]//Proceedings of the IEEE 10th Annual Ubiquitous Computing,Electronics & Mobile Communication Conference.New York:IEEE,2019:280-285.
[20] LUO P,ZHU Z,LIU Z,et al.Face Model Compression by Dis-tilling Knowledge from Neurons[C]//Proceedings of the AAAI Conference on Artificial Intelligence.AAAI,2016:3560-3566.
[21] GE S,ZHAO S,LI C,et al.Efficient Low-Resolution Face Re-cognition via Bridge Distillation[J].IEEE Transactions on Image Processing,2020,29:6898-6908.
[22] ZHANG K,GE S,SHI R,et al.Low-Resolution Object Recognition with Cross-Resolution Relational Contrastive Distillation[J].IEEE Transactions on Circuits and Systems for Video Technology,2024,34(4):2374-2384.
[23] HAO Z,GUO J,HAN K,et al.One-for-all:bridge the gap between heterogeneous architectures in knowledge distillation[C]//Proceedings of the 36th Conference on Advances in Neural Information Processing Systems.PMLR,2024:123-135.
[24] HUANG Z H,YANG S Z,LIN W,et al.Knowledge Distil-lation:A Survey[J].Chinese Journal of Computers,2022,45(3):624-653.
[25] ZHAO B,CUI Q,SONG R,et al.Decoupled Knowledge Distillation[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition.IEEE,2022:11953-11962.
[26] JI M,HEO B,PARK S.Show,Attend and Distill:KnowledgeDistillation via Attention-Based Feature Matching[C]//Proceedings of the AAAI Conference on Artificial Intelligence.2021:7945-7952.
[27] PENG B,JIN X,LIU J,et al.Correlation Congruence forKnowledge Distillation[C]//Proceedings of the IEEE International Conference on Computer Vision.IEEE,2019:5007-5016.
[28] ZHU Y,WANG Y.Student Customized Knowledge Distillation:Bridging the Gap Between Student and Teacher[C]//Procee-dings of the IEEE International Conference on Computer Vision.IEEE,2021:5057-5066.
[29] YANG C,AN Z,CAI L,et al.Knowledge Distillation Using Hierarchical Self-Supervision Augmented Distribution[J].IEEE Transactions on Neural Networks and Learning Systems,2024,35(2):2094-2108.
[30] HU J,SHEN L,SUN G.Squeeze-and-Excitation Networks[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition.IEEE,2018:7132-7141.
[31] KRIZHEVSKY A,NAIR V,HINTON G.Learning MultipleLayers of Features from Tiny Images[R].Toronto:University of Toronto,2009.
[32] SUN Y,WANG X,TANG X.Deep Learning Face Representation from Predicting 10,000 Classes[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition(CVPR).IEEE ,2014:1891-1898.
[33] HUANG G B,MATTAR M,BERG T,et al.Labeled Faces in the Wild:A Database for Studying Face Recognition in Unconstrained Environments[C]//Proceedings of the Workshop on Faces in ‘Real-Life’ Images:Detection,Alignment,and Recognition.Marseille:INRIA,2008.
[34] GUNTHER M,HU P,HERRMANN C,et al.Unconstrained Face Detection and Open-Set Face Recognition Challenge[C]//Proceedings of the IEEE International Joint Conference on Biometrics.IEEE,2017:697-706.
[35] ZAGORUYKO S,KOMODAKIS N.Wide Residual Networks[J].arXiv:1605.07146,2016.
[36] ZHANG X,ZHOU X,LIN M,et al.ShuffleNet:An Extremely Efficient Convolutional Neural Network for Mobile Devices[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition.IEEE,2018:6848-6856.
[37] SANDLER M,HOWARD A,ZHU M,et al.MobileNetV2:Inverted Residuals and Linear Bottlenecks[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition.IEEE,2018:4510-4520.
[38] TIAN Y,KRISHNAN D,ISOLA P.Contrastive Representation Distillation[C]//Proceedings of the International Conference on Learning Representations.2020:1-19.
[39] HEO B H,KIM J,YUN S D,et al.A Comprehensive Overhaul of Feature Distillation[C]//Proceedings of the IEEE International Conference on Computer Vision.2019:1921-1930.
[40] PARK W,KIM D,LU Y,et al.Relational Knowledge Distil-lation[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition.2019:3967-3976.
[41] SIMONYAN K,ZISSERMAN A.Very Deep Convolutional Networks for Large-Scale Image Recognition[J].arXiv:1409.1556,2014.
[42] WANG H,WANG Y,ZHOU Z,et al.CosFace:Large MarginCosine Loss for Deep Face Recognition[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition.2018:5265-5274.
[43] MENG Q,ZHAO S,HUANG Z,et al.MagFace:A UniversalRepresentation for Face Recognition and Quality Assessment[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition.2021:14225-14234.
[44] KIM M,CHOI J,KIM D,et al.AdaFace:Quality adaptive margin for face recognition[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition.IEEE,2022:18750-18759.
[45] LI P,TU S,XU L.Deep Rival Penalized Competitive Learning for Low-Resolution Face Recognition[J].Neural Networks,2022,148:183-193.
[46] WANG Z,CHANG S,YANG Y,et al.Studying Very Low Re-solution Recognition Using Deep Networks[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition.2016:4792-4800.
[47] SINGH M,NAGPAL S,SINGH R,et al.Dual directed capsule network for very low resolution image recognition[C]//Proceedings of the IEEE International Conference on Computer Vision.2019:340-349.
Viewed
Full text


Abstract

Cited
  Shared   
  Discussed   
No Suggested Reading articles found!
渝ICP备16013121号-4
地址：重庆市两江新区洪湖西路18号    邮编：401121  
电话：023-63500828（编辑部） 023-67039612（会议/专题） 023-67039623（财务/发票）
E-mail：jsjkx12@163.com
本系统由北京玛格泰克科技发展有限公司设计开发