Abstract: Independent component analysis (ICA), mainly based on single linear functions, is an approach widely used in face recognition. A new approach to face recognition—kernel independent component analysis(KICA) was introduced that was based on an entire function space of nonlinear subspace. First introduced ICA in a concise way and mainly discuss the KICA’s basic principle and algorithm, and analyzed the differences between the ICA and KICA in face recognition. Finally, the experimental and analytical results show that in face recognition KICA algorithm outperforms ICA algorithm.

Key words: kernel independent component analysis (KICA), independent component analysis (ICA), kernel generalized variance (KGV), face recognition

摘要： 在人脸识别中提出一种基于非线性子空间的核独立成分分析(KICA)方法。在简单介绍了ICA方法的基础上,对KICA方法的基本原理和算法作了较为详细的描述。为了验证基于KICA和ICA的人脸识别方法的识别效果,进行了对比实验和分析。实验和分析结果表明,在人脸识别中,基于KICA的方法优于基于ICA的方法。

关键词: 核独立成分分析, 独立成分分析, 广义核方差, 人脸识别