针对非均衡数据的人脸识别算法  

Implicit Semantic Data Enhancement for Unbalanced Data in Face Recognition

作　　者：周辉[1] ZHOU Hui(College of Computer&Communication Engineering,China University of Petroleum,Qingdao 266580)

机构地区：[1]中国石油大学(华东)计算机与科学技术学院,青岛266580

出　　处：《计算机与数字工程》2023年第2期277-281,490,共6页Computer & Digital Engineering

基　　金：国家自然科学基金项目(编号:61379106,61379082,61227802);山东省自然科学基金项目(编号:ZR2013FM036,ZR2015FM011);中央高校基本科研业务费专项资金(编号:18CX06046A)资助。

摘　　要：目前在基于深度学习的人脸识别领域面临数据不均衡的问题,这使得样本少的类别精度难以提升。常见的通过对损失函数的改进,或是简单的数据增强难以从根本上解决这一问题。为此,论文将隐式数据增强的方法引入基于角距离度量函数,构建人脸识别算法:通过对样本在特征空间中的特征向量沿某个特定方向做相关改变来增加此类样本的多样性,使样本较少的类别在特征空间中尽可能地产生更多相似却不相同的特征向量来减轻不均衡数据集对识别算法所造成的负面影响。在通用人脸识别数据集LFW,IJB-A,MegaFace上的测试结果表明,该方法能有效提高人脸识别的精度。At present,face recognition based on deep learning faces the problem of data imbalance,which makes it difficult to improve the accuracy of categories with few samples.It is difficult to solve this problem fundamentally through the improvement of loss function or simple data enhancement.To this end,the paper introduces implicit data enhancement method to angular distance measurement function:face recognition algorithm based on the sample feature vector in the feature space along a specific direction make relevant changes to increase the diversity of samples,the samples less categories in the feature space as far as possible to produce more similar but not identical feature vector to reduce unbalanced data sets to identify the negative impact caused by the algorithm.The test results on the universal face recognition datasets LFW,IJB-A and MegaFace show that this method can effectively improve the accuracy of face recognition.

关 键 词：神经网络 人脸识别 隐式数据增强 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]