论文部分内容阅读
Fisher判别分析是统计模式识别中经典的有监督维数约简方法,可以在最大化类间散度的同时最小化类内散度,但存在分析过程中仅使用有标记数据而忽略无标记数据的问题.鉴于此,提出基于概率类和不相关判别的半监督局部Fisher(SLFisher)方法,以实现半监督学习的高维映射到低维的类间数据对尽可能地分离,且类内邻近数据尽可能地紧凑.采用2组标准数据集进行实验,结果表明了SLFisher方法能够有效提高识别率.
Fisher discriminant analysis is a classic supervised dimension reduction method in statistical pattern recognition, which can minimize divergence within classes while maximizing the divergence between classes. However, there is a problem that only the labeled data is used in the analysis and the unlabeled data is ignored In view of this, a semi-supervised local Fisher (SLFisher) method based on probability class and uncorrelated discriminant is proposed to achieve the best possible separation of high-dimensional semi-supervised learning from low-dimensional class data, The data is as compact as possible.Experimental results using two sets of standard data sets show that the SLFisher method can effectively improve the recognition rate.