A new map element extraction method for color cartographic maps was developed. Most studies on cartographic maps have been restricted to binary or gray scale maps. However, many map elements in Korea are colored differently to be recognized easily. It is natural to use their color information for extracting them.
The proposed method is divided by three major sections: pre-processing, crude color segmentation, and map elements extraction.
The pre-processing transforms the RGB (red, green, and blue) color map image scanned by optical color scanner into an HSV (hue, saturation, and value) color coordinate image. The colors of the map elements are selected for easy identification. This can be achieved by leaving enough space in Hue or Saturation or Value. Thus, the HSV color coordinates are better than the RGB color coordinates in color map segmentation.
A crude global thresholding technique is used for the next process. All maps have only several colors to express each map object differently. This permits crude segmentation. Global thresholding is best for rough segmentation.
The process starts with segmenting the hue component image. Maps usually have only three hues. These are separated widely in hue level and can be divided easily. Main hue values must be input before the process. Using the values, the hue image is segmented by a valley detection method in hue histogram. Because the hue values can be changed in another map, we designed the program to require the main hue value inputs.
After hue segmentation, the process continues in the saturation and the value components. The colors of the map elements are widely separated in saturation or value planes to be recognized easily. For example, main streets are colored with a dark red: a high saturation and low value. City areas are tinted with a light red: a high saturation and high value. Buildings are colored with black: a low saturation and low value. Thus, a simple threshold scheme is effective to segment the image. We choose threshold levels by the average of the maximum and the minimum values in the saturation and the value components, and finally, can get seven separated regions: high saturation and high value red region, high saturation and low value red region, low saturation and high value red region, low saturation and low value red region, high value green region, low value green region, and blue region.
Post-processing beautifies the segmented map elements by their shape and obtains the correct map elements. Each element has its own color and shape characteristics.
Main streets can be obtained from high saturation and low value region. Contour lines, unfortunately, are drawn by a dark red the same as that of main streets. The differences between the two are the thickness of the lines and the black boundaries. Main streets are delineated by two parallel black lines. After including these black lines, the correct main streets were obtained by the width of regions.
The crude city area can be obtained from the high value and low saturation regions. As the main streets, calculating the width of each object, the correct city area should be extracted.
Black regions are very important in maps. They represent railroads, border lines, names, buildings, etc. The black can be obtained from low value regions. Unfortunately, they contain the contour lines, segments with low value and fairly high saturation. The main difference between the two is the value of saturation component. Contour lines have a fairly high saturation value. Thresholding the saturation components, we can get both: the black regions and the contour lines.
The blue regions can be obtained easily from the first hue separation process.
We proposed a new method for segmenting color cartographic maps. This method uses the color and the geometry information of map elements. This can be the first process of a fully automated color map mapping system.
In Chapter III, this thesis includes a new threshold selection method for separation of objects and background. This method used a human separation method for objects and background, examined the changes in the number of foreground regions in the image with varying threshold and determined a proper threshold value. The proposed technique was applied to several images and was compared with other histogram-based segmentation methods. The possiblilty of extension of multi-level thresholding was also examined.