中文摘要：

栅格数据矢量化是遥感和地理信息系统一体化集成的关键技术之一。在以往的研究中，对“岛”，尤其是多层嵌套岛的处理还不够完善。本文提出了一种新的矢量化方法，其步骤如下：（1）提取端点和结点。使用2×2模板匹配窗口对图像进行一次遍历，提取出端点和结点以及它们之间的连接信息；（2）形成弧段。利用端点和结点的连接信息来形成非闭合弧段和闭合弧段；（3）建立多边形。对于闭合弧段在某种意义上就是一个简单多边形，只需将弧段的结点转换为多边形的结点就可形成由它组成的简单多边形，对于非闭合弧段，则使用左转算法来形成由它们组成的简单多边形；（4）对“岛”进行处理。由于可能发生一个多边形同时被多个多边形包含的情况，为了建立正确的拓扑关系，需要对这种包含关系进行分析，找到岛的最小包含多边形。实验证明，本算法较好地对“岛”的问题进行了处理，并且速度较快，能够满足实际应用的要求。

英文摘要：

Raster data vectorization is one of the key techniques for the integration of remote sensing and GIS. Although it＇ s a traditional problem with a lot of algorithm proposed, there still exist many defects including, e. g. , the problem of the ＂island＂, especially the multi-nested ＂island＂ in many vectorizing methods. In this paper, an improved algorithm for raster data vectoriation is proposed. This method includes four main steps. The first step is to extract the vertices and nodes. Moving through the image using a 2 x 2 template window, we can extract the vertices and nodes and their connection information. The second step is to construct the arcs. We use the connection information to construct the non-closed and closed arcs. The third step is to build the polygons. With left-turn algorithm, we can build simple polygons from non-closed arcs. Besides, since closed arcs correspond to simple polygons, we only have to convert a vertex of an arc to a vertex of a polygon. The last step is to deal with the ＂islands＂. We judge containing relationship among these polygons. Since a polygon may be contained by several parent polygons, we should find its minimal containing parent and then deal with the ＂islands＂. The main advantages of the proposed algorithm include the follows. 1 ） Algorithm efficiency is relatively fast. 2） It isn＇ t sensitive to the size of raster data, but only sensitive to the number of the nodes and vertexes. 3） It can build the correct topology and then solve the problem of multi-nested ＂island＂. Our method is validated with many successful experiments when applied in vectorizing the remote sensing classification imagery.