一种融合WPCA与WLDA的人脸识别方法

Combination of WPCA and WLDA for Face Recognition

提出了一种融合加权PCA与加权LDA的人脸识别方法。该方法将主成分分析(PCA)的优点和线性鉴别分析(LDA)的优点充分地融合在一起,以欧式距离为参数的权函数w1使得PCA降维时加入类别差异从而得到最优投影矩阵,解决了PCA过程中使用最小距离方法时识别精度相对低的缺点;以马氏距离为参数的权函数w2使得LDA分类时,进一步扩大类间离散度,减小类内离散度。另外该方法在识别精度上比WPCA+LDA、PCA+WLDA、PCA+LDA算法都有很大的提高,通过在ORL、AR、FERET人脸库上的实验验证了算法的有效性。

A face recognition algorithm combining WPCA and WLDA is developed. The proposed algorithm integrates the merits of PCA and LDA. Weight functionw; which based on Euclidean distance makes it add class differences in the PCA dimension reduction, then the optimal projection matrix is obtained. It can overcome the PCA＇ s shortcomings of lower precision when using the minimal distance method. And in the LDA classification weight functionw2 based on Mahalanobis distance further expands the between-class scatter and reduces the within-class scatter. Moreover,this method improves the recognition accuracy greatly compared with the WPCA ＋ LDA,PCA ＋ WLDA and PCA ＋ LDA. Many experiments on ORL, AR and FERET face database indicate that our algorithm is effective.