基于分层关系度量网络的单样本人脸识别  

Single-sample Face Recognition Based on Hierarchical Relationship Measurement Network

作　　者：周晓康 钟锐[1] 宋亚峰 ZHOU Xiaokang;ZHONG Rui;SONG Yafeng(School of Mathematics and Computer Science,Gannan Normal University,Ganzhou 341000,China)

机构地区：[1]赣南师范大学数学与计算机科学学院,江西赣州341000

出　　处：《赣南师范大学学报》2024年第3期35-41,共7页Journal of Gannan Normal University

基　　金：国家自然科学基金(62266003);江西省自然科学基金资助项目(20232BAB202056);江西省教育厅科技项目(GJJ211401);江西省基础教育研究课题(SZUGSZH2021-1147)。

摘　　要：单样本训练集中的每个类只有一张样本,训练样本数量的不足将使模型得不到有效训练,使得模型无法准确提取具有类内变化的人脸面部特征,导致模型的识别率低下.针对该问题,提出了一种基于分层关系度量网络(Hierarchical Relation Measure Network,HRMN)的单样本人脸识别模型.首先,使用语义网络将训练集中人脸进行性别层次的语义划分;随后,应用小批量K均值聚类算法对所划分的第一层语义人脸特征进行分层聚类,得到具有多个不同抽象层次的分层特征树(Hierarchical Feature Tree,HFT).最后,使用所构建的多层关系度量网络计算出不同抽象层次面部特征与目标样本间的加权融合相似度,根据相似度得出目标样本的类别信息.为了验证所提算法的有效性,本文进行了大量的实验,实验结果表明,该模型优于几种近年来应用较为广泛的单样本人脸识别模型.There is only one sample for each class in the single-shot training set,and the insufficient number of training samples will make the model unable to be effectively trained,making the model unable to accurately extract the facial features with intra-class variations,resulting in a low recognition rate of the model.To solve this problem,a single-sample face recognition model based on Hierarchical Relation Measure Network(HRMN)was proposed.Firstly,the semantic network was used to divide the training focus on the face of the gender level.Subsequently,the small-batch K-means clustering algorithm was used to classify the first layer of semantic face features to obtain a hierarchical feature tree(HFT)with multiple different abstract levels.Finally,the weighted fusion similarity between facial features at different levels of abstraction and the target samples was calculated by using the constructed multi-layer relationship measurement network,and the category information of the target samples was obtained according to the similarity.In order to verify the effectiveness of the proposed algorithm,a large number of experiments are carried out,and the experimental results show that the proposed model is superior to several single-sample face recognition models that have been widely used in recent years.

关 键 词：聚类算法 分层分类 单样本学习 单样本人脸识别 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]