结合分形理论和修改的方向最大最小方差的居民地提取方法

Fractal Feature and Direction Variances(FFDV):A New Algorithm for Urban Area Extraction from Remote Sensing Image

传统的居民地提取方法存在以下不足:第一是提取的居民地存在误判的情况;第二是提取的居民地边界不准确,效率低;第三是很难适用于不同空间分辨率的图像。为克服这些问题,本文提出了一种基于分形理论和方向最大方差(DMAV)和修改的方向最小方差(MDIV)的居民地提取方法,算法共分为三个步骤:第一步是利用分形特征和修改的方向最小方差提取居民地的候选区域;第二步是利用形态学滤波和面积阈值法除去干扰区域,并对提取的居民地区域进行修整;最后是利用方向最大最小方差,对居民地边界进行精确的定位。本文用了大量的遥感图像验证了提出算法,实验证明,本文方法是精确的、有效的。

Conventional methods for urban area extraction have several shortcomings： Firstly, there are false detected regions. Secondly, the edge of extracted urban area is not precise. The third is that they can not fit to images with different resolutions. To overcoming these disadvantages, we proposed a new algorithm based on fraetal, direction maximal variance （DMAV） and mended direction minimal variance （MDMIV） to extract urban areas exactly. It is divided into three steps： Firstly fractal and MDMIV are used to detect candidate urban regions. Secondly, morphologic and area-threshold are used to eliminate eyelets and get rid of the disturbing small regions. Finally, we use DMAV and MDMIV to find the accurate edge point positions. Lots of experiment results prove that the proposed algorithm is valide and accurate.