基于K-L变换和平均近邻法的人脸识别

首先介绍了K-L变换和近邻法判别的原理;然后以MATLAB R2009b为实验平台,分别利用类间散布矩阵和总体散布矩阵作为K-L变换的的产生矩阵,对ORL人脸库的400幅图片进行K-L变换,一部分作为训练样本,一部分作为待识别样本,训练样本以产生特征脸空间;接着计算出待识别图片在特征脸空间的坐标,采用平均近邻法进行人脸识别.最终的实验结果给出了基于两种产生矩阵的算法时间和正确识别率,实验证明采用K-L变换对人脸提取特征很有效,本文基于K-L变换和平均近邻判别法的人脸识别的方法正确率最高可达到95％.

Face recognition based on K-L transform and average-nearest neighbor rules

In this paper, I firstly introduced the K-L transform and the nearest neighbor rules for discrimination; then I performed an experiment based on Matlab R2009b as follow： Conduct K-L transform on 400 photos that from the ORL face database using the Between-Class Scatter Matrix and the Total-Scatter Matrix respectively as the generation matrix. All the samples divided into two parts, as a sample to be recognized and a training sample which used to generate the eigenface space. Then I compute the coordinates of the photos to be recognized in the eigenface space and finally use the Average-nearest neighbor rules for face recognition. The experiment result includes correct recognition rate and the time taken by program. The result show that using K-L transform and Average-nearest neighbor rules for face recognition is effective, and we can reach up to 95% accuracy of recognition.