The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B3b. Beijing 2008 
high. The computation time problem reduced the practical 
attractiveness of this method and the alternative method pf 
increasing ellipse was followed. 
The threshold value for coincide path determination was chosen 
to be 0.2. It should be mentioned that this value is selected 
based on the S/N ratio of the input road raster map. Figures 9 
and 10 show the obtained result of road key point identification 
for two input images of Figures 6a and 7a. in this images, each 
road patch with its representative ellipse is presented in a 
separate colours. 
Figure 9. Road key point identification of input image 6a 
Figure 10. Road key point identification of input image 7a 
A fuzzy shell clustering was performed based on the 
predetermined coincide road patches and vague road pixels 
were determined. These pixels are shown in Figures 9 and 10 
with yellow colour.Finally, considering the centroid of vague 
road pixels, the road connections were generated as shown in 
Figures 11 and 12 for input source images of Figures 6a and 7q 
respectively. 
i 
Figure 12. Road vectorization of input image 7a 
In order to evaluate the performance of road vectorization 
procedure, three accuracy assessment parameters were designed 
and computed which are called “Mean Deviation”, 
“Completeness” and RMSE. 
Mean deviation is computed as follows: 
Mean Deviation = 
Deviation Area 
Road Legth 
(6) 
Where, deviation area is the area between the identified road 
centerline and the manually reference extracted road vector map. 
Completeness, as the second vectorization accuracy assessment 
parameter, represents the length percent of the extracted road 
network in the input image. 
Table 2 shows the obtained accuracy assessment parameters for 
both sample#! and Sample #2. 
Mean 
Deviation 
Max. 
Deviation 
Completeness 
RMSE 
Sample #1 
0.53 
2.00 
88% 
0.55 
Sample #2 
0.38 
1.41 
84% 
0.49 
Table 2. Accuracy assessment parameters of road vectorization 
procedure 
554