XXXIX-B4, 2012 
s tracked sequentially 
L. 
'oduce several terms in 
( pixel in the 3 X: 3 
0 black pixels in the 
in three black pixels in 
au" 
(c) 
int. € Crossing point 
lescribed as follows. 
he starting point in a 
the sliding window. 
ck pixels in the 3X3 
tion d to the current 
a human operator). 
set it to be white), and 
is counted from 1) . 
hood. There are three 
nains only one black 
nvisible point, use its 
03. 
t is a true or pseudo 
the following). If it is 
yrward direction d, go 
ie sliding window, go 
g points. If it is less 
' centre of the sliding 
pi. 
be encountered due to 
ures, or due to noise. 
iile the latter is called 
ossing point is met, 
it, and the next point 
racking continues as 
S met, an automatic 
to across the pseudo 
concentration degree 
imber of black pixels 
the point in the un- 
' crossing point P. If 
r than a preset value, 
it. Otherwise it is a 
solution of 300 dpi, 
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 
the concentration degree can be set as 20 within a 5 X 5 region. 
When meeting a pseudo crossing point, a trial-tracking is done 
to determine the next tracking direction. As shown in Figure 8, 
there are two forward directions d/ and d2 at the pseudo 
crossing point P. First, three connecting pixels along d/ and d2 
are tracked respectively, and each of their corresponding grey- 
scale values in the grey-scale image are recorded. Then, the 
average grey-scale values are calculated for d/ and d2, 
respectively. If the former is smaller, d/ is determined as the 
next tracking direction, else d2. 
  
  
  
  
  
  
  
Figure 8. Determination of the forward direction at a pseudo 
crossing point 
Figure 9 illustrates the continuous sliding window and 
sequential line tracking from the starting point with an initial 
direction pointed by the arrow. Figure 10 shows the results of 
image segmentation, thinning and line tracking in window 1-6 
in Figure 9a. The grey line marked in each window draws a 
tracking path. By connecting all the grey lines in order, the 
tracking result is obtained (see Figure 9b). 
  
(a) (b) 
Figure 9. Sequential line tracking. a Continuous sliding 
window. b The result of line tracking 
  
  
  
  
  
  
  
(c) 
Figure 10. a Images in window 1-6 in Figure 9a. b 
Corresponding segmentation results of current linear feature. € 
Corresponding results of thinning and line tracking 
  
4. EXPERIMENTS AND ANALYSIS 
Experiments have been conducted to test our proposed method. 
Figure 11a is a part of a topographic map with relief shadings. 
The size of the image is 300 X 300 pixels, the resolution is 300 
dpi, and the sliding window is 25 X 25 pixels. For each contour 
line, once a starting point and direction are input by a human 
operator, it can be tracked automatically. In the case that an 
intersection or a gap is met, automatic tracking stops, and a new 
point in the front is input manually. After that, the line tracking 
continues. If the gap is wider, a few points should be collected 
manually to get over it, and then automatic tracking resumes. 
Figure 11b is the vectorization result of contour lines. Figure 
12a is a part of another topographic map with forest tints. The 
image size, the scanning resolution, and the window size remain 
unchanged. Figure 12b is the vectorization result of contour 
lines. The average time of vectorizing contour lines in Figure 
11 and Figure 12 are 200 seconds and 160 seconds respectively 
on a 3 GHz Pentium (R) 4 computer. Most of the time was 
taken by manual input of starting points and some interventions 
during the tracking process, while the time required by 
automatic tracking is negligible. 
  
  
  
  
Figure 11. A part of a scanned colour topographic map with 
relief shadings. a Original image. b Vectorization result of 
contour lines 
  
  
  
  
  
Figure 12. A part of another topographic map with forest tints. 
a Colour scanned image. b Vectorization result of contour lines 
A mass of other maps including colour, grey and black-and- 
white maps have been vectorized, and satisfactory results have 
been achieved. The experiments shows that the proposed 
method is of great practical value in vectorizing linear features 
directly in original topographic map images especially those 
colour maps with forest tints and relief shadings. 
Furthermore, a comparison with commercial software MapGIS 
has been made. For scanned colour maps with clear contrast 
103 
E 
oe ges tis pea 
2M nn ds