中文摘要：

为减弱离群点对数据处理的影响,提出了一种鲁棒的加权核主成分分析算法。利用核函数将样本投影到核空间,在核空间构建一个样本加权重建误差最小模型,最大限度地提取数据中的非线性信息并降低离群点样本的干扰。在Yale人脸库和UCI数据集上的实验表明,该方法具有很好的识别率,尤其对离群点样本具有较好的鲁棒性。

英文摘要：

This paper proposed a robust weighted KPCA to reduce the effect of outliers in data processing.By introducing kernel function to project samples into kernel space,it constructed a model minimizing weighted reconstruction errors in the kernel space,to maximize nonlinear information extracted from the data and reduce the interference of outlier samples.Experiments on Yale face database with outliers and UCI data sets show that the proposed method has better recognition rates and robustness especially with outliers.