Basic Study on Automatic Drawing Input System and Evaluation Method for Recognition Results

自动绘图输入是建立地理信息系统（GIS）和绘图数据库系统的关键技术。根据正在开发和广泛使用的几种基于商业的系统，用户希望对这些系统进行比较以选择更好的系统。然而，目前在图纸输入领域还没有被广泛接受的评价标准。本研究的目的之一是建立一种标准的绘图输入系统之间的比较方法，另一个目的是推进自动绘图输入系统。为了比较计算机系统之间的性能，人们提出并广泛使用了几种基准测试。在本研究中，提出了基准图，以便于图输入系统的相互比较。图输入的目的和图的性质与图的用途和图的种类是不同的。在本研究中，选择地图来提出评价方法的必要性。这个对象的地图在本研究中被称为基准地图。在选择地图时，考虑了地图包含各种地理标志和描绘典型地貌的条件。为基准测试选择的映射如下：(1) 1/25000地图OJIYA的一部分，(2)1/25000地图TOKYO-SYUBU的一部分，(3)1/2500地图TOTTORI的一部分。(1)、(2)选取地图上15cm × 12cm的部分，(3)选取地图上23cm × 18cm的部分。对答案向量数据进行手工数字化处理，提出了一种基于基准图和答案数据的评价方法。本研究的另一个课题是提出一种快速、稳健的绘图矢量化方法。针对这一目标，提出了一种结合核心线提取和细化的矢量化方法。然后，通过实验验证了该方法的优点。

Automatic drawing input is an essential technique to build geographic information systems (GIS) and drawing databased systems. According to several commercial based systems being developed and widely used, the users wants to compare these systems to chose better one. The evaluation standard widely accepted, however, dose not exist on the field of drawing input, until now. One of the objecta of this research is to establish a standard method for comparison among drawing input systems, and another object is to advance automatic drawing input systems.To compare the performance among computer systems, several kinds of benchmark tests are proposed and widely used. In this study, benchmark drawings are proposed for the sake of mutual comparision of drawing input systems.The purpose of the drawing input and the nature of the drawings are quit different from the applications and the kind of the drawings. In this research, map is selected to propose the necessity of the evaluation method. The maps for this object are called benchmark maps, in this research.On the selection of the maps, the condition which the maps include various geographical symbols and depicted typical configuration of the ground were considered. The maps selected for benchmark are the followings ; (1) a part of 1/25000 map OJIYA, (2) a part of 1/25000 map TOKYO-SYUBU, (3) a pat of 1/2500 map TOTTORI.(1) and (2) are selected 15cm x 12cm part on the maps, and (3) is selected 23cm x 18cm part of the map. The answer vector data were made by hand-digitizing, and an evaluation method used benchmarks maps and the answer data proposed.Another topic of this research is to propose a fast and robust drawing vectrization method. For this object, a vectorization method using a combination of core-line extraction and thinning is proposed. Then, the good features of the proposed method are presented by experiments.