基于动态主成分子空间的人脸识别算法

Face Recognition Based on Dynamic Principal Component Subspace

在基于子空间分析的人脸识别中,通常是按照特征值的大小来确认主成分的重要性,并以此为基础构造一个固定的特征子空间。通过人脸图像重建分析,发现固定的特征子空间会给人脸识别带来误差,于是采用多元线性回归分析理论,提出一个动态主成分子空间构造算法。在此基础上,得到了动态PCA(主成分分析)算法和基于Gabor特征的动态PCA算法。由ORL和Georgia Tech人脸数据库上的实验结果表明,该算法不仅减少了主成分数目,而且提高了识别率。

The significance of principal component was determined by the corresponding eigenvalue in face recognition based on subspace analysis, then a static feature subspace was established. However,it could result in an inaccurate performanee by analyzing the process of face reconstruction. A dynamic feature subspace algorithm was proposed according to multiple linear regression analysis. Furthermore, a dynamic principal component analysis （DPCA） and a Gabor feature based dynamic principal component analysis （GDPCA） were brought forward. Experiment results on ORL and Georgia Tech face databases show that the proposed algorithm not only decrease the number of principal components but also increase the correct rate of face recognition.