一种新的线性判别准则在人脸识别中的应用

针对Fisher线性判别分析(LDA)在进行人脸识别这种小样本问题分类时常常遇到类内散度矩阵Sw奇异,而无法直接应用的问题,提出一种新的线性判别准则,即:定义一个新的准则函数,在对类内散度矩阵无奇异性要求的情况下,找到此准则函数最优的权向量.应用此判别准则和Fisher LDA方法分别在ORL人脸数据库上选取40个人的图像进行识别率的测试,Fisher LDA方法的识别率为0.95,而本文方法的识别率可以达到0.955,优于Fisher LDA方法.

Application of a new linear discrimination criterion to human face recognition

Aimed at the problem of the singularity of in-category divergence matrix Sw constantly encountered in the classification of such a mini-sample as human face identification with Fisher's linear discrimination analysis(LDA) that cannot be directly applied in such case,a new linear discrimination criterion was proposed,where a new criterion function was defined and the optimal weighing vector in this function was found as well.A test for image identification efficiency of forty persons selected from ORL human face database was performed respectively with the discrimination criterion presented and Fisher's LDA method.It was found that the identification efficiency was 0.95 with the latter and,with the former,it amounted to 0.955 that was superior to the Fisher's LDA method.