基于域自适应均值网络的素描人脸识别方法  

SKETCH FACE RECOGNITION BASED ON DOMAIN ADAPTATION MEAN NETWORK

作　　者：陈长武 曹林[1,2] 郭亚男 杜康宁 Chen Changwu;Cao Lin;Guo Yanan;Du Kangning(Key Laboratory of the Ministry of Education for Optoelectronic Measurement Technology and Instrument,Beijing Information Science and Technology University,Beijing 100101,China;School of Information and Communication Engineering,Beijing Information Science and Technology University,Beijing 100101,China)

机构地区：[1]北京信息科技大学光电测试技术及仪器教育部重点实验室,北京100101 [2]北京信息科技大学信息与通信工程学院,北京100101

出　　处：《计算机应用与软件》2023年第4期107-115,共9页Computer Applications and Software

基　　金：国家自然科学基金项目(61671069);北京信息科技大学“勤信人才”培育计划项目(QXTCPA201902);北京市教委面上项目(KM202011232021);校基金-基于对抗学习的素描人脸识别研究项目(2025017)。

摘　　要：针对素描图像和光学图像间模态差异大,以及传统深度学习方法在少量训练数据情况下易过拟合的问题,提出一种基于域自适应均值网络的素描人脸识别方法。该方法设计元学习训练策略将学习水平从数据提升至任务,来提升模型的泛化能力;提出一种均值损失来辅助特征提取器提取判别性特征;在训练集的素描图像域和光学图像域之间引入一种域自适应模块来减少二者模态差异。在UoM-SGFS素描人脸数据库和e-PRIP素描人脸数据库上进行实验,结果表明该方法优于其他算法。Aimed at the problem that there is large modal difference between sketch images and optical images,and the traditional deep learning methods are prone to over-fitting with a small amount of training data,a sketch face recognition method based on domain adaptive mean network is proposed.The meta learning training strategy was designed to improve the learning level from data to tasks,so as to improve the generalization ability of the model.A mean loss was proposed to assist the feature extractor in extracting discriminant features.A domain adaptive module was introduced between the sketch image domain and the optical image domain of the training set to reduce their modal differences.Experiments on UoM-SGFS sketch face database and e-PRIP sketch face database show that the proposed method is superior to other algorithms.

关 键 词：素描人脸识别 过拟合 元学习 域自适应 

分 类 号：TP3[自动化与计算机技术—计算机科学与技术]