摘要
该文针对实用人脸识别中的光照、表情、姿态等变化问题,通过因子分析和数据挖掘提出一种鲁棒的人脸识别方法。本文首次提出基于因子分析的人脸识别方法,并分析基于内容与风格信息的因子分析模型的人脸识别方法与基于Fisher线性判别分析的人脸识别方法的一致性。为了提高该方法的鲁棒性,通过两因子方差分析与加性模型分离人脸内因子与外因子,降低风格信息对人脸观察特征的影响。实验结果表明:此方法比Fisher脸方法具有更高、更稳健的性能,特别是在Fisher脸方法无能为力的复杂环境下能表现出较好的性能。
Considering the variation of illumination,expression and pose,a new face recognition algorithm is proposed based on factor analysis and data mining.The consistence of factor analysis model based on content and style with linear discriminant analysis in face recognition is analyzed.In order to improve the robustness of this method,two-factor analysis of variance and additive model is proposed to reduce the impact of style information on face observed feature.Experimental results show that this method has higher and more stable performance than "Fisherface" method.Especially,when the "fisherface" method performance is bad under complex environments while this method demonstrates better performance.
出处
《电子与信息学报》
EI
CSCD
北大核心
2011年第7期1611-1617,共7页
Journal of Electronics & Information Technology
基金
中央高校基本科研业务费专项资金(20102120103000004)
郑州市重大科技攻关项目(072SGZS38042)资助课题
关键词
人脸识别
因子分析
线性判别分析
加性模型
Face recognition
Factor analysis
Linear discriminant analysis
Additive model