基于子空间分析的人脸识别方法研究

Research of face recognition methods based on subspace analysis

人脸识别技术是模式识别和机器视觉领域的一个重要研究方向,基于子空间分析的特征提取方法是人脸识别中特征提取的主流方法之一。本文对目前应用较多的子空间分析方法进行了研究,具体介绍了线性子空间分析方法:主成分分析(PCA)、线性鉴别分析(LDA)、独立主成分分析(ICA)、快速主成分分析(FastICA),和非线性子空间分析方法:基于核的PCA(KPCA)的基本思想以及这些方法在人脸识别中的研究进展和一些新的研究成果。此外,还应用orl及YaleB人脸库对几个基础的子空间方法进行了验证实验。实验结果表明,在几个子空间分析方法中,FastICA算法取得了最高的识别率。最后,结合实验结果对各算法的优缺点进行了分析总结。

Face recognition is an important research direction of pattern recognition and machine learning. Among many approaches to the face recognition, the feature extraction methods based on subspace analysis give the most promising results, and have become one of the most popular methods. In this paper, subspace analysis methods were research and several kinds of the linear subspace method, such as Principal Component Analysis（PCA）, Linear Discriminant Analysis （LDA）, Independent Component Analysis （ICA）, Fast ICA and nonlinear subspace methods such as Kernel PCA（KPCA） were introduced. The basic principles and their research achievements of these methods were deseriped and the analysis applications to face recognition were given. Moreover, the ORL and YALE B databases were used to verify these basic subspace methods. The experiment results indicate that FastlCA method is more powerful than other subspace methods for face recognition. Finally, the advantages and disadvantages of these methods were demostrated by dicussing the experimented results.