【作者】 辛动军；

【导师】 周献中；

【作者基本信息】 南京理工大学 ， 控制科学与工程， 2007， 博士

【摘要】 地理信息的数字化是建立地理信息系统(GIS)的重要环节，工作量占整个系统开发的三分之二以上。地理信息系统中数据的获取仍是影响其发展的瓶颈。彩色地图要素的自动识别与获取技术涉及到多个学科，是集理论和实践为一体的图像处理与识别技术。多年来的理论及实践为此课题的深入研究奠定了良好的基础，但也存在着许多亟待解决的问题。本文以比例尺为1：5万的彩色地形图为研究对象，重点对颜色分层、等高线的识别与获取、黑板要素图上道路的识别与获取、蓝版要素图上水系的识别与获取几个方面进行了研究。利用本文提出的算法，结合本课题已有研究成果开发了一个彩色地形图识别与获取原型系统。本文的主要研究成果及创新点：1．提出了基于类中心约束的模糊c均值(fuzzy c-means，FCM)聚类算法。根据RGB空间中颜色的统计特征初始化隶属度矩阵，解决了传统FCM算法对初始值敏感的问题。为了克服FCM算法不能很好的处理聚类尺寸不同和数据疏密程度不同的情况，定义了新的度量方式。除利用了颜色信息外，还引入了像素关系信息，从而有效地降低了颜色误差对分色的影响。2．提出了等高线跟踪算法和基于等高线线体流向分析的补断算法。利用数学形态学的击中—击不中变换消除等高线图上的噪声和孔洞。为了提高等高线的矢量化效果，克服噪声的影响，采用改进的梯度矢量流主动轮廓模型提取等高线。在未经细化的等高线图上直接提取等高线，可避免因等高线细化畸变导致的跟踪错误。利用等高线线体流向信息并结合等高线邻接关系修复断裂的等高线。3．提出了虚线道路和实线道路的提取算法。构造了基于格式塔准则的虚线道路图搜索A~*算法启发式函数。定义了提取实线道路基元的投影矩阵。改进了提取实线道路的主动轮廓模型，提高了对弱边缘的提取能力。4．提出了不同水系要素的不同提取方法。包括不同交叉模式的跟踪方法、双线河流的判别方法、改进的梯度矢量流模型及面状要素提取的初始化方法等。5．根据以上思路和算法，开发了一个彩色地形图矢量化原型系统。更多还原

【Abstract】 Geographic information digitalization, which takes about two thirds of thewhole workload, is one of the key tasks in Geographical Information Systemdevelopment. In digital image processing and recognition, automatic interpretationand acquisition of topographic map is a challenging work both in theory and practice,and is thus a bottleneck of the development for the GIS system, because it involvesmany subjects. Previous theories and applications have provided a solid foundationfor further research, but there are still some urgent and thorny problems.This dissertation takes color topographic map with a scale of 1:50000 as theresearch object, Algorithms for automatic recognition and vectorization of thetopographic map, including color segmentation, vectorization of contour lines, roadsand water system elements are proposed. And a prototype system for automaticrecognition and acquisition of topographic map elements is developed by utilizingthese algorithms and integrating other research achievements of our research team.Main contributions of the dissertation include:1. Fuzzy c-means (FCM) segmentation algorithm of color topographic map with theconstraints of cluster center is proposed.To solve the initialization of FCM, the pattern matrix is initialized according tostatistical color characterization in RGB space. Segmentation result of fuzzy c meansalgorithm is improved according to the data density by avoiding coherent clustercenters and combining spatial relation information.2. The contour line track algorithm and the rehab algorithm of broken lines arepresented.Mathematics morphological hit-miss operations are used to remove isolatedpixels and holes within the contour lines. To attain good vectorization result, animproved gradient vector flow active contour model is explored to track contourlines. To avoid tracing error caused by thinning distortion, contour line is extractedon unthinned map. To rehab broken and cohering contour lines, region directionalfield is adopted3. Extraction algorithms of dashed road and solid road are proposed, includingHeuristic function of A~* search algorithm extracting dashed road based on Gestalt rules is proposed. Defining projection matrix to extract solid road primitives. Solidroad is extracted utilizing improved active contour model.4. Different extracting methods are presented for different water systemelements, These methods including a track method for various intersection models, amethod to distinguish double-line river elements, an improved gradient vector flowmodel and an initialization method to extract area-like water system elements.5. Based on the thoughts and algorithms proposed above, a prototype system forautomatic recognition and acquisition of color topographic map elements isdeveloped. The validity of proposed algorithms have been confirmed withexperiments更多还原