Abstract: There exists the small sample problems in the traditional Fisher discriminant method, and the recognition rate of the inverse Fisher method is also relatively low. So this paper proposes an inverse Fisher face recognition method based on kernel function, which uses a proper kernel function to extract some effective face features in the high dimensional space. Experimental results show that the new method keeps the robustness of the inverse Fisher discriminant as well as the high recognition rate of faces.

Key words: face recognition, inverse Fisher discriminant, feature extraction, kernel function

摘要： 传统Fisher判别方法存在小样本问题，而逆Fisher判别方法的识别率较低。为此，提出一种基于核函数的逆Fisher人脸识别方法，在逆Fisher准则的基础上引入核函数映射，选取合适的核函数在高维空间里提取人脸图像特征。实验结果表明，该方法能保持逆Fisher判别的鲁棒性，人脸识别率较高。

关键词: 人脸识别, 逆Fisher判别, 特征提取, 核函数

CLC Number: 

• TP391