Semi-supervised fuzzy learning strategy by using a way of partitioning the outlier instances
离群样本划分的半监督模糊学习策略

In this paper, a semi-supervised fuzzy learning algorithm based on the partitioning of the outlier feature space is presented. First, a reformative fuzzy LDA algorithm using a relaxed normalized condition is proposed to achieve the distribution information of each sample represented by a fuzzy membership degree, which is incorporated into the redefinition of the scatter matrices. Moreover, we approach the problem of parameter estimation by considering the formulation of the Hopfield neural network. Using this method, the first key step of the fuzzy classification is addressed. Second, considering the negative influences from the outlier instances, we separate the outliers from the whole feature space by means of the distribution information of each sample. The strength of the technique is that it successfully uses the improved fuzzy supervised algorithm as a feature extraction tool, while quantifying those factors that exert influence ons the outlier class assignment, by means of the fuzzy semi-supervised method. Extensive experimental studies conducted on the NUST603, ORL, XM2VTS and FERET face image databases show that the effectiveness of the proposed fuzzy integrated algorithm.