International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B3. Istanbul 2004 
starting parameter and these measurements are given to the 
algorithm by the operator. (Vosselman, 1998). 
These kind of algorithms have snake algorithm basis. Snake 
algorithm uses the energy function and aims at the result 
product of the two methods mentioned above. 
This method has some difficulties: 
i. Relating to the radiometric quality of the image, the 
alteration threshold of pixels in gray values are not 
constant. 
il. Because of the noise in the image, some of the road- 
making pixels have different gray values and this 
causes broken contours. 
iii. In urban areas, objects on the roads may cause the 
image of the road to appear broken. 
The Algorithm developed in this study is principally a guide 
algorithm. However, it uses the gray values belonging to the 
contours that the operator chooses as the starting parameter. 
2.3.1. Determining the Colour Threshold of Contours: 
This Process is used to define raster to vector conversion 
parameters. The parameters are calculated by training contour 
samples. Except for this, the different ground colors on the map 
effects the contour pixels color directly. Thus, with this step, a 
new training group is built for each ground color and different 
threshold values are calculated for each group. 
The collected data is the gray value of the pixels belonging to 
their RGB bands. For calculation process, minimum and 
maximum RGB values are found for each group. 
R() € Map() 
G() e Map() (1) 
B() € Map() 
Roe RO RS € R( 
G x € 00 Cus € CQ (2) 
B, € PO, Bua € BO 
where RO, GO, BO = bands of scanned map image. 
Rao Gao Bg ^ maximum gray values of the contour 
sample set. 
Rain Gai» Bag ? maximum gray values of the contour 
sample set. 
2.3.2. Raster to Vector Conversion: To start the automatic 
elevation value election process, the image matrix is scanned 
from left to right. If the scanned pixel and at least one of its 
eight neighbors are in the gray value of contours, they are 
chosen as the starting pixels. This x, y image coordinated pixel 
is taken as the i, j coordinated central element of 3x3 nucleus 
matrix. If the gray values of pixels in Ryiy- Rx, Gai Gg, and 
Bmmin-Bmax (Eq. 2) then it is given a value of 1 and the others 0. 
The center of gravity is calculated using the coordinates of 
pixels that have 1 gray value. The binary pixels are created like 
Costa (2001). This value is taken as a central pixel of the new 
nucleus matrix. If the centers of gravity of binary image pixels 
intersect with the nucleus matrix central pixel, the closest I 
gray value pixel is chosen as the next target. 
Thus, the first matched pixel is used as the starting point of the 
polygon. The recent pixel coordinate is added to the polygon. 
Again the recent pixel is taken as the center pixel of nucleus 
matrix. For these steps, gray values of pixels which are in the 
elevation value threshold are taken as 1 and the others are taken 
as 0. While calculating their center of gravity, the central pixel, 
the neighbors of the former central pixel and nucleus matrix are 
not taken into consideration. The process is repeated in this 
way. 
2.33. The Node Elimination Process of the Vectors: After 
the vector data production process, approximately 500 000 to 
2 000 000 broken contours occur, according to the topological 
structure of the map. This abundance of broken contours 
requires a large memory and increases the time spent for the 
analyses. The increase in the analyses time is geometrically 
proportional with the number of broken contours. 
Line generalization algorithms are not used for decreasing the 
number of broken contours, because the broken points 
mentioned are in the middle of the pixels and they make an 
angle of 45 or multiples of this number. These kind of sharp 
angles between broken contours cause the line generalization 
algorithms to increase the number of broken points. 
  
  
  
Figure 1. Vectors from raster (pink colored) 
In this study, a method is chosen to delete the broken contours 
systematically. The only parameter of the method is the number 
of the nodes to be deleted. This choice is made by the user 
interactively, according to the topological structure of the map 
(Figure 1). 
2.34. Editing the Broken Contours: Elevation lines can 
be broken because of some deformations or other cartographical 
objects on the map. The greatest factor that affects the speed of 
this step is the number of broken contours. Another important 
parameter is the threshold value distance for connections to be 
chosen, since the elevation lines are very close especially in 
hilly areas. If a wide threshold value is chosen, neighbor curves 
can be joined. So, it is concluded in this study that this choice 
should be made by the operator. 
462 
  
  
Internati 
2.4. 
This proc 
the map 
this proce 
Intel Pen 
this proc 
results ii 
accelerate 
used for 
done by ( 
For this 
length (ir 
hand, th 
threshold 
too. Then 
2.8. 
Due to t 
wrong co 
elevation 
The steps 
2.6. 
The user 
values. TI 
When the 
very few. 
according 
A satisfac 
the lines 
software, 
given ele 
be delete