融合Gabor与局部切空间排列法的人脸识别算法

摘要

针对Gabor小波提取人脸特征存在维数高,计算复杂的问题,引入基于划分的局部切空间排列算法(Partitional Local Tangent Space Alignment)对得到的Gabor幅度特征(Gabor Magnitude Feature,GMF)进行降维,同时将主成分分析(PCA)和线性判别分析(LDA)引入到算法中,确定用最近邻分类器进行分类识别的最优投影子空间.通过在ORL人脸数据库上的实验证明了该算法的有效性,用Gabor小波提取特征对光照和表情变化等有良好的鲁棒性.%Using the Gabor wavelet to extract the face features has the problems of high dimensions and complex calculations. To solve the problems, Partitional Local Tangent Space Alignment (PLTSA) is introduced as a method to operate on Gabor Magnitude Features (GMF) to extract the submanifolds. In the same time PCA and LDA are introduced in the method to find the best projection subplace in which the nearest distance classifier is used for classification. Experiments with ORL database show that the approach is quite effective. This method also has good robustness with the changes of light and expression because of using gabor wavelet.