核主成分分析与典型相关分析相融合的人脸识别

FACE RECOGNITION BY FUSING KERNEL PRINCIPAL COMPONENT ANALYSIS AND CANONICAL CORRELATION ANALYSIS

人脸识别具有小样本、高维等特性。典型相关分析算法(CCA)无法准确提取人脸识别特征,不能准确刻画人脸图像的局部变化,导致人脸识别率低。为提高人脸识别率,提出一种核主成分分析与典型相关分析相融合的人脸识别算法(KPCA-CCA)。首先将人脸图像划分多个子模块,然后提取局部特征,同时采用KPCA提取全局特征,并采用CCA将两种特征进行融合,降低特征向量的维数,最后采用子模式进行人脸识别,以投票方式确定人脸的类别。采用AR与Yale数据集对KPCA-CC性能进行测试,仿真结果表明,相对于对比模型,KPCA-CCA提高了人脸识别的识别率。

Face recognition has the features of small sample and high-dimensionality. Canonical correlation analysis （CCA） can＇ t accu- rately extract the features of face recognition, nor accurately depicts the local variations of face image as well, which lead to low face recogni- tion rate. In order to improve the face recognition rate, in this paper we propose a novel face recognition algorithm which fuses kernel princi- pal component analysis and canonical correlation analysis （KPCA-CCA）. First, it divides the face image into multiple sub-models and ex- tracts local features, meanwhile the KPCA is employed to extract global features, and then these two kinds of features are fused by CCA to re- duce the dimensionality of eigenveetors, finally the sub-models are used for face recognition, and the face type is determined by voting. The performance of KPCA-CCA has been tested by AR and Yale datasets, simulation results show that it raises face recognition rate with respect to the reference model.