逆C均值学习样本筛选方法

An Inverse C-mean Method for Filtering the learning Samples

在有监督分类中,歧义学习样本将导致学习时间增长,分类性能下降等问题。本文提出一种逆C均值样本筛选方法,可有效剔出歧义的训练样本。该方法采用类似C均值聚类分析逆过程的方法,将有歧义的学习样本从训练集中剔除。本文以有监督图像分割中的样本选择为例进行实验。实验表明,该方法可有效消除由人机交互引入的歧义样本。

The error samples may lead to long time and low performance in the classification using the supervised learning. In this paper, an Inverse C-Mean method was proposed to filter the error samples. The proposed method is like the inverse procedure of C mean cluster. Experiments illustrated that the mentioned method was valid to filter the error sample in the learning samples.