用小波变换和Fisher判别对人脸进行特征提取

Feature extraction of human face using kernel Fisher discriminant

提出了一种用小波变换和核函数Fisher判别对人脸进行特征提取的方法.同传统的特征提取方法相比,用核函数Fisher判别进行特征提取,不仅可以对人脸图像进行维数压缩,而且还可以有效利用提样本的类别信息.同时,用小波变换对人脸图像进行预处理以降低计算复杂度.同传统的Fisher变换相比,可以较好地解决人脸识别这一非线性问题.实验结果表明方法是有效的.

In this paper, a method employing the kernel Fisher discriminant and wavelet transform to complete the feature extraction for human face recognition is proposed. Compared with several commonly used methods for feature extraction, the proposed method can not only process dimension reduction, but also provide infor- mation for classification. Furthermore, it performs well in linearly nonseparable case. So optimal results can be achieved for human face recognition, which is a nonlinear problem. To reduce the computational complexity, the wavelet transform is applied to the pretreatment of original human face images. The experiments on ORL dataset prove the efficiency of the proposed method.