摘要： 结合主元分析(Principal Components Analysis, PCA)与线性鉴别分析(Linear Discriminant Analysis, LDA)的特点，提出用于性别鉴别的PCA-LDA算法。该算法通过PCA算法求得训练样本的特征子空间，并在此基础上计算LDA算法的特征子空间。将PCA算法与LDA算法的特征子空间进行融合，获得PCA-LDA算法的融合特征空间。训练样本与测试样本分别朝融合特征空间投影，从而得到识别特征。利用最近邻准则即可完成性别鉴别。基于ORL(Olivetti Research Laboratory)人脸数据库的实验结果表明，PCA-LDA算法比PCA算法识别性能好，在性别鉴别中是一种有效的方法。

关键词: 性别鉴别, PCA-LDA算法, 融合算法

Abstract: Combined with the advantages of principal components analysis (PCA) and linear discriminant analysis (LDA), PCA-LDA on gender classification is presented. Feature sub-space of training samples is obtained by way of PCA, and feature sub-space from LDA is calculated on the basis of PCA. In the meanwhile, the two feature sub-spaces from PCA and LDA are fused, and the fusion feature space is acquired. After training samples and test samples are respectively projected towards the fusion feature space, recognition features are accordingly gained. Nearest neighbor rule is utilized in gender classification. Experimental results on ORL face database show that PCA-LDA is better than PCA in recognition performance, and is a valid method in gender classification.

Key words: Gender classification, PCA-LDA algorithm, Fusion algorithm