摘要
本文给出了一个无监督图像类别聚类的新方法,该方法基于信息理论原理-信息瓶颈。本聚类方法基于阶段性的分组:首先,对给定文档中的每一张图像应用高斯混合模型,在选定的特征空间中,以一组相连接的区域来表示图像。然后,确保簇和图像内容之间的互信息最大化,对图像进行分组。簇的合适数量可直接由信息瓶颈原理决定。实验结果显示出了该聚类方法在真实图像数据库中的表现。
A new non-monitored image clustering method was proposed in this paper, which is based on the information bottleneck, This method is classified based on the steps. Firstly, the Gaussian mixture model was applied to each image in a certain documents, in which, the collective regions are used to present the images. And then, the images were grouped in order to maximize tile cross information between the content of cluster and image. The right number of clusters can be determined by the iuformation bottleneck, The performance of this method was approved by the exoeriments at last.
出处
《微计算机信息》
北大核心
2008年第24期308-309,189,共3页
Control & Automation
关键词
图像类别
无监督聚类
图像分组
信息瓶颈
Images Category
Non-monitored Cluster
Image Grouping
Gaussian mixture model