基于独立分量分析的普适人脸识别系统

在模式识别领域,变量间的高阶统计关系开始受到更多关注.但目前许多人脸识别系统一方面依赖二阶统计关系,另一方面又需先采用主分量分析技术对样本进行降维.主分量分析技术自身却对二阶统计关系敏感,因此需要寻找一种对高阶统计关系敏感的算法作后续处理.为此作者提出了一种基于独立分量分析的普适人脸识别系统,并与传统的基于Fisher线性判别规则的人脸识别系统进行了比较分析,重点讨论在光照方向大幅度变化和人脸图像不完整情况下两种系统性能的优劣.理论分析和实验结果均证实,在这两种情况下,基于独立分量分析的普适人脸识别系统的性能优于传统的基于Fisher线性判别规则的人脸识别系统的性能.

Generalized Face Recognition System Basing on Independent Component Analysis

In pattern recognition area,higher-order statistical relationship among variables attracts mare and more attention.Most current face recognition systems depend on second-order statistical relationship,which require principal component analysis to reduce dimensions of samples.But principal component analysis itself also depends on second-order statistical relationship,so an algorithm depending on higher-order statistical relationship is needed.In this paper we propose a generalized face recognition system basing on independent component analysis,and compare it with traditional system based on fisher's linear discriminant under large variation of lighting direction and with incomplete face images.Theoretical analysis and experimental results show that the new generalized face recognition system could outperform traditional system basied on fisher's linear discriminant under large variation of lighting direction and with incomplete face images.