关键词: 人脸识别, 独立成分分析法（ICA）, 核主成分分析（KPCA）, 核独立成分分析（KICA）, 特征选择

Abstract:

       We employ improved k-nearest neighbor Relief algorithm to select features after extracting features with kernel independent component analysis (KICA) to make the extracted independent elements to be favorable to be recognized. This improved Relief algorithm can reduce noise pollution and address the issue of small samples, so the selected features can be employed to better classify faces. We prove the effectiveness of this method with the experiment to AR face database and comparisons with feature selection algorithms of intra- and inter-cluster distance.

Key words: face recognition;independent component analysis, kernel principal component analysis, kernel independent component analysis, feature extraction