利用Mahalanobis距离进行人脸表情的识别

Mahalanobis distance based facial expression recognition

提出了利用Mahalanobis距离进行人脸表情识别的方法．首先将待分类的图像样本集进行坐 标变换,使得变换以后类间离散度尽可能大而类内离散度尽可能小,即使变换以后的Fisher准则函数 取得极大值,在新的坐标下求每个待分类样本到各类均值向量的Mahalanobis距离,从而将待分类的 样本归到Mahalanobis距离最小的类中去,通过实验得到了平均80．25％的识别率．

It was proposed that Mahalanobis distance based technique could classify four facial expressions namely： neutral, joyful, surprised and sad. First we＇ve mapped all the sample vectors to a new coordinate by using the optimal linear transform matrix W that makes Fisher criterion function maximum. Then in this new coordinate we calculated the Mahalanobis distance of every sample vector to the mean vectors of the four classes and classified it to the class to which its Mahalanobis distance is the least, and the average recognition rate we＇ve got is 80.25%.