多元线性回归在非负矩阵分解人脸识别中的应用

Application of Multivariable Linear Regression in NMF Face Recognition

单一次优非负基特征蕴含的人脸分类信息有限,分类精度受制于基特征的低维表示。针对非负矩阵分解(NMF)的弱分类特点,分析NMF的人脸识别过程,提出增加多组基特征扩充可用弱分类信息;利用多组弱分类标签向量的相关性,采用多元线性回归方法建立集成标签映射,整合关联的弱类别信息,在次优基特征低维表示基础上提取正确的类结构关系。利用多种人脸数据集的实验结果表明,符合统计属性的集成标签映射能显著改善NMF的人脸识别能力。

Single sub-optimal nonnegative basis features usually contain limited face category information and the recognition rate lies on the related low dimensional representation.On account of the weak classification of NMF,more basis features were proposed to develop more latent correlated category information by looking at the face recognition process of NMF carefully.And then,the multivariable linear regression methods were used to build label mapping from ensemble weak labels to true label.It integrated the weak correlated category structure information and rose the correct category structure to the surface well.The results on several face databases show that the statistical label mapping enhances the face recognition capability of NMF.