考虑局部邻域多流形度量的单训练样本人脸识别  

Single Training Sample Face Recognition Algorithm Considering Local Neighborhood Multi Manifold Metric

作　　者：毛明扬 MAO Mingyang(School of Date Science,Guangzhou Huashang College,Guangzhou 511300)

机构地区：[1]广州华商学院数据科学学院,广州511300

出　　处：《计算机与数字工程》2022年第7期1562-1565,1572,共5页Computer & Digital Engineering

基　　金：国家自然科学基金项目(青年基金)“非线性特性多智能体系统一致性研究及其应用”(编号:61403219);广州华商学院校内导师制科研项目(编号:2022HSDS07)资助。

摘　　要：针对单训练样本中识别干扰因素较多的问题,为了增加人脸识别效果,在考虑局部邻域多流形度量的基础上,提出一种新的单训练样本人脸识别算法。首先预处理人脸图像信息,将影响因素降到最低;划分数据集,根据划分结果对分布在流形结构内数据点计算近邻点;构建多流形距离度量矩阵和误差度量矩阵;将人脸图像经过投影降维转换为低维流形结构,完成单训练样本人脸识别。实验结果验证了所设计算法识别效率较高和平均识别率较高,所用时间较少,具有很好的优势。Aiming at the problem that there are many interference factors in single training sample recognition,in order to in⁃crease the face recognition effect,a new single training sample face recognition algorithm is proposed based on the local neighbor⁃hood multi-manifold measurement.Firstly,the face image information is preprocessed to minimize the influencing factors.The data set is divided and the nearest neighbor points are calculated according to the partition results.The multimanifold distance measure matrix and error measure matrix are constructed.The face image is transformed into a low-dimensional manifold structure through projection reduction,and the single training sample face recognition is completed.Experimental results show that the proposed algo⁃rithm has higher recognition efficiency,higher average recognition rate and less time,which has certain advantages.

关 键 词：局部邻域多流形度量 人脸图像 数据集 近邻点 距离度量矩阵 

分 类 号：TN346[电子电信—物理电子学]