基于线性投影分析和FCM动态聚类的人脸识别

Face recognition based on linear projection analysis and FCM dynamic clustering

依据主成分分析方法(PCA)对图像具有很好的表达能力,即能很好地重构原图像,而线性鉴别分析(LDA)可使图像样本具有较高可分性的特点,提出对图像先进行PCA处理,再进行LDA处理,从而降低人脸特征维数并对人脸图像进行了特征提取;并提出用FCM动态聚类算法作为识别分类器,对人脸进行识别。实验和分析结果表明,在人脸识别中,这种融合PCA和LDA的分类方法能够更好地对特征进行提取,且FCM动态聚类分类器比K近邻判别分类器更具有灵活的分类能力。

This paper proposed a face recognition classifier based on FCM dynamic clustering. First, as for linear projection, principle component analysis and linear discriminant analysis were expounder. Where PCA seeked directions that were efficient for representation, LDA seeked directions that were good at discriminating samples. Then, used FCM dynamic clustering calculate threshold to recognize face. Experiments on ORL face database used the combining algorithms above to extract features. The experimental result indicates that the recognition performance of classifier combination in decision level is superior to others, and is more robust.