基于Fisher准则的半监督特征提取方法  

Semi-supervised feature extraction method based on Fisher criterion

作　　者：郝伟[1] 刘忠宝[2] 

机构地区：[1]山西工商学院计算机信息工程学院,山西太原030006 [2]中北大学计算机与控制工程学院,山西太原030051

出　　处：《计算机工程与设计》2017年第1期238-241,共4页Computer Engineering and Design

基　　金：山西自然科学基金项目(201601D011042);山西省高等学校科技创新基金项目(2014142)

摘　　要：针对实际应用中得到的数据往往只有少量具有类别标签,大多数类属未知的情况,在Fisher准则的基础上,提出基于Fisher准则的半监督特征提取方法 SFEM。在构造邻接图的基础上,重新定义类内离散度矩阵和类间离散度矩阵,利用Fisher准则找到的最优投影方向满足类间离散度矩阵与类内离散度矩阵之比最大,保证样本能较好地分开。若干标准数据集上的仿真结果表明,SFEM在解决半监督特征提取问题上具有一定优势。Mass unlabeled data and a small quantity of labeled data exist in practice. To fully utilize the labeled and unlabeled data, semi-supervised feature extraction method based on Fisher criterion （SFEM） was proposed based on the depth analysis of the traditional semi-supervised feature extraction methods. The adjacent graph was constructed, and the within class scatter matrix and the between-class scatter matrix were redefined. Fisher criterion was used to ensure the samples in different classes apart from each other. Comparative experiments on several standard datasets verify the effectiveness of SFEM in solving the problem of semi-supervised feature extraction.

关 键 词：特征提取 半监督算法 费希尔准则 类内离散度 类间离散度 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]