Abstract
摘要 在本論文中，我們發展一個讀取城市街道圖訊息的系統。此系統包含色彩分離、符號辨識、道路選取以及相關文字字串的分割。我們使用K-means來分離顏色，並且發現此方法具有良好的效果。利用傳統的和改進的Hough Transform找出地圖的經線與緯線，並且依照地圖經緯線的距離以及旋轉角度，來判別地圖的大小和旋轉方向；因而在使用樣型比對做符號辨識時，可以增加準確度。同時我們也使用黑色圖素密度，來辨別與符號相關的文字位置。道路的選取則採用資料結構的概念，依此概念選取出單一的路。我們提出一個能有效萃取地圖訊息的系統。Abstract In this thesis, we introduce a map reading system toextract all the information in the city maps. The systemconsists of color segregation,symbol recognition, roadextraction, and related character strings extraction. We use K-means to separate colors, and it works effectivelyfor the mapimages. Conventional and improved Hough transform havebeenproposed in extracting longitudes and latitudes. Templatematching has also been developed to recognize symbols which areadjusted by the orientation of the longitudes and latitudes inthe map. Than we apply the black pixel density to identify therelated characters of symbols. Road extraction applies an ideaof data structure. In the experiments, we illustrate that oursystem can extract all the information in the city mapeffectively.