基于马尔科夫网视频图像目标检测跟踪方法研究

Study on application, approach and concept of DMN in segmentation of images

研究了马尔科夫网的概念、方法,分析它在空间数据挖掘中的作用与意义,并以视频图像为例广泛研究以多种粒度(节点数)建立视频图像间的马尔科夫网,通过网络结构分析检测视频图像中的目标差异。研究表明马尔科夫网可以很好地揭示数据间的抽象近邻关系,并且这种网络自身就具有表达知识的意义。

The paper studies concept and method of DMN, analyses its role in spatial data mining, advances new notion of DMN's grainess, and deeply investigates two score metrics (the maximum posterior probability and minimum builtup entropy) which are used in constructing DMN in order to segment images, and explores two algorithms of global and local hunting within DMN for segmentation of images. The researches and experiments indicate that the DMN may apropos reveal abstract adjacent relations existed in data; and the DMN oneself has capabilities of data mining and showing knowledge. Lastly, we point out future works of its application and continue studying.