基于多流形的单样本人脸模糊分类算法  

Multi-manifold based fuzzy classification algorithm for single face image problem

作　　者：徐洁 杨长茂 陈建平[1] 王文琰 XU Jie;YANG Chang-mao;CHEN Jian-ping;WANG Wen-yan(Computer Science,Guangzhou Maritime University,Guangzhou 510725,China;Faculty of Automation,Guangdong University of Technology,Guangzhou 510006,China)

机构地区：[1]广州航海学院计算机学院,广东广州510725 [2]广东工业大学自动化学院,广东广州510006

出　　处：《计算机工程与设计》2025年第3期719-725,共7页Computer Engineering and Design

基　　金：国家自然科学基金项目(61773128);广东省普通高校重点领域专项基金项目(2023ZDZX3018);广州市教育规划课题基金项目(202316988);广州市黄埔区2024年度一般课题基金项目(2024Y3);2024年广州市教育局高校科研基金项目(2024312392)。

摘　　要：为解决单个人脸样本分类中样本数量不足的问题,提出一种多流形模糊分类算法(FMMC)。通过分割图像增加“样本”数量,构造类别子流形。引入模糊集理论,定义类别流形隶属度,弱化不同类别子流形上语义相同的图块相似度,强化同一类别子流形上不同位置图块的类别信息相关性,有效限制离群图块对分类结果的影响,提高分类的性能。在3个公开人脸数据库上进行实验,其结果表明,FMMC对单个样本问题的分类可行且有效。To solve the problem of insufficient number of samples in the classification of individual face samples,the fuzzy multi-manifold classification algorithm(FMMC) was proposed.The number of samples was increased by segmenting the image,and the category sub-manifolds were constructed.The fuzzy set theory was introduced to define the affiliation degree of category manifolds,which weakened the similarity of semantically identical blocks on different category manifolds,and strengthened the relevance of category information of blocks at different locations on the same category manifold,effectively limiting the influence of outlier blocks on the classification results,and improving the performance of classification.Experiments were conducted on three public face databases.Experimental results show that FMMC is feasible and effective for classification of single sample problems.

关 键 词：单样本 K最近邻分类器 模糊集 多流形 切割 流形隶属度 分类 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]