ne XXXIX-B4, 2012 
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Volume XXXIX-B4, 2012 
XXII ISPRS Congress, 25 August — 01 September 2012, Melbourne, Australia 
VECTORIZATION OF LINEAR FEATURES IN SCANNED TOPOGRAPHIC MAPS 
USING ADAPTIVE IMAGE SEGMENTATION AND SEQUENTIAL LINE TRACKING 
Yun Yang ^*, Xiaoya An^, Limin Huang? 
* State Key Laboratory of Geoinformatic Engineering, Xi'an Research Institute of Surveying and Mapping, Xi'an, 
710054, China - 
yytoall@126.com 
KEY WORDS: Colour map, Vectorization, Sliding window, Image segmentation, Line tracking 
ABSTRACT: 
Vectorization of cartographic features is a difficult task in geographic information acquisition from topographic maps. This paper 
presents a semi-automatic approach to linear feature vectorization directly in original scanned colour topographic maps without 
layer separation. In the approach, a sliding window is added on a user-given linear feature, and the current line in the window is 
segmented adaptively by using colour space conversion, k-means clustering and directional region growing. A thinning operation is 
then performed in the window and the line is tracked from the centre to the edge. By moving the window continuously along the line, 
vectorization is carried out by iterative operations of image segmentation, thinning and line tracking. Experimental results show that 
linear features in colour topographic maps can be vectorized rapidly and accurately, especially in those maps with forest tints and 
relief shadings. 
1. INTRODUCTION 
Map digitization is one of the most important means of 
geographic data acquisition for Geographic Information System 
(GIS) applications. In the past decades, map digitization has 
undergone the stage of manual tracking on digitizing tablets and 
the stage of heads-up screen digitizing. Currently, human- 
machine interactive digitization based on image processing and 
pattern recognition techniques has been more and more used. 
Many commercial systems such as VPStudio, RxAutoImage, 
R2V and MapGIS are available for map digitization. 
Topographic maps are vectorized from raster images into vector 
data by raster-to-vector conversion. For black-and-white line 
drawings with high quality, automatic vectorization can be 
implemented by using line tracking method. However, for 
colour topographic maps, where cartographic features with 
different colours overlap and intersect one another, automatic 
vectorization is still a challenging task. 
The existing techniques of colour map vectorization can be 
grouped into two main categories: colour segmentation based 
and original map based. The colour segmentation based method 
first split the colour map image into several layers with 
predefined map colours, and then vectorize cartographic 
features in the separated binary layers interactively or 
automatically. This kind of method can significantly reduce the 
complexity of cartographic features, and therefore make map 
vectorization relatively easy. Here, colour segmentation is a 
fundamental step. The precision and speed of map vectorization 
rely on the quality of segmented images. Many algorithms for 
colour segmentation including statistic pattern classification 
(Feng et al., 1996; Su et al., 1999), neural network (Lin et al., 
1994; Guo et al., 1998), fuzzy clustering (Zheng et al., 2003), 
and others have been developed. Nevertheless, these algorithms 
are far from being satisfactory. Due to the influence of original 
paper map's quality and the scanner's performance, colours in a 
scanned map image are not the same as that in the original map. 
Large numbers of scattered colours and noises exist in the 
  
* Corresponding author. 
image, especially for the map with forest tints and relief 
shadings. So it is difficult to segment the colour map image into 
perfect colour layers with high qualities. Noises, gaps and 
adhesions appear everywhere in the separated layers, 
particularly in the contour line layer. Although some algorithms 
for removing noises and connecting broken lines have been 
developed (Cheng et al., 2003; Zeng et al., 2004), a great deal 
of human editing is still inevitable to acquire high quality image 
for automatic vectorization. 
Another way to vectorize linear features is based on original 
scanned colour maps. In this method, a linear feature is tracked 
starting from a user-specified point, and some kinds of flexible 
user interventions are allowed in case where automatic tracking 
fails. This approach utilizes the strategy of human-machine 
cooperation, makes the line tracking process under human 
control, and provides the ability to correct data immediately if 
required. Therefore, it is more practical for colour map 
vectorization, and can be used as a complimentary of the former 
method. It should be a preferred one for the map image with 
poor quality. Unfortunately, to the best of our knowledge, very 
few research works have been done on such vectorization 
method (Wu et al., 1998; Huang et al., 2005). 
MapGIS is one of the most powerful GIS software systems 
produced by Zondy Cyber Group Co., LTD, which provides the 
way of interactive line tracking in original colour map images. 
In the process of line tracking, the minimum colour distance 
rule is adopted to determine the next tracking point. From the 
point of view of applications, this method has two main 
disadvantages in performance: 
(1) The line tracking process depends greatly on the user- 
specified starting point. The user needs to magnify the image 
and select the midpoint of a line exactly as the starting point. If 
the selected point has a little deviation, the following line 
tracking cannot be ensured to be along the centreline. 
99