改进PCA-LDA的人脸识别算法研究

Research on improved face recognition algorithm of PCA-LDA

针对主成分分析(PCA)未充分利用样本类别信息及线性鉴别(LDA)在小样本时识别率低的问题,提出了一种基于PCA和LDA相融合的人脸识别算法。该算法首先将输入人脸图像进行几何位置规范化和灰度分布均衡化预处理;然后利用PCA算法将人脸训练图像投影到低秩特征子空间,利用LDA算法计算类间离散度大、类内离散度小的特征子空间,从而获得PCA-LDA的人脸融合特征空间;最后将训练样本、测试样本投影至融合的特征空间,并利用最近邻准则实现对测试样本的识别。实验结果表明,该算法能够有效融合PCA和LDA的优势,提高系统识别的鲁棒性和效率。

To solve the problems of that principal component analysis(PCA)underutilized sample’s classification information and linear discriminant analysis(LDA)low recognition rate in small sample,a PCA and LDA face recognition fusion algorithm is proposed.Firstly,the input face image is preprocessed with the geometric position normalization and the gray level distribution equalization.Then the low-rank feature subspace of training face image is obtained by the way of PCA,and then the subspace face feature of large dispersion of the between-class and small dispersion of the within-class from LDA is derived.Thus the fusion feature is acquired.Finally,the training and test samples are projected to the fusion feature space,and the nearest neighbor criterion is used to identify the samples.The experimental results show that the algorithm can effectively combine the advantages of PCA and LDA,and can improve the system recognition robustness and efficiency.