基于Bayes网络的航空图象理解模型

An Aerial Image Understanding Model Based on Bayesian Networks

提出一种基于Bayes网络的图象理解模型,用于检测与描述城市航空图象中的停车场、空地、房屋等二维和三维目标.模型首先使用一种基于感知组织的算法生成焦点区域,然后使用Bayes网络建立焦点区域的图象特征与目标之间的(不确定性)联系,由Bayes网络推理机制完成对焦点区域的识别.对于屋顶区域,模型使用一种基于DEM(digitalelevationmap)方向直方图的匹配算法进行模板匹配,生成房屋的三维几何模型描述.模型中所有Bayes网络组成层次结构,可以方便地增加被识别目标的种类;图象特征与提取图象特性的算法之间由一种称为证据源的数据结构相连接,可以方便地扩充图象特征的提取算法.由此实现了模型的易扩展性.实验结果表明了本文模型的有效性.

A Bayesian network-based image understanding model is presented,which can be used in detection and describing two and three dimensional objects such as parks,grounds and buildings in aerial image of cities.The focused region is first extracted in the model by a perceptual-organization-based algorithm.Then the uncertainty relationships between image features of the focused region and objects are represented in Bayesian networks.Finally,the focused regions are recognized through the inference engine of the Bayesian networks.As to the roof of the building,a DEM based direction histogram is used as the matching algorithm to template matching before the generation of description of the three dimensional geometric model.All Bayesian networks in the model are organized in a hierarchy structure,which can increase with ease the variety of objects to be recognized.The image features and their extraction algorithms are connected by a data structure called evidence source,which expands the extraction algorithms easily,thus increasing the extendability image-understanding model .The experimental results have shown the validity of this model.