Yandong Wang 
  
4 TESTS AND RESULTS 
An aerial image in Hunter valley, New South Wales was used in this study. The original image has a scale of 1:25,000 
and has been scanned with a pixel size of 30 um. To generate a low-resolution image, the original image was resampled 
at a reduction factor of 15, resulting in an image of 500x500 pixels with a ground resolution of 11.25 m, giving an 
image with similar resolution to SPOT panchromatic images. The image covers a rural area with two main roads, as 
shown in Figure 2a. A morphological operator was used in this study to extract line features. After line extraction, line 
points are tracked and small gaps (< three pixels) are bridged to form elongated line segments based on the criterion of 
collinearity. Smooth line segments were generated using a split-and-merge operation. Three different thresholds of 5, 10 
and 20 pixels were used to generate the hierarchy of line images as shown in Figure 2b, c and d. As can be seen, the line 
image using the threshold of 5 pixels contains most road parts and many non-road line features which are very short 
(Figure 2b). Due to blurred road surface around the middle and lower right of the image, roads in these areas were not 
extracted correctly by the line operator. With the increase of the threshold, more non-road line features were removed, 
but some road segments were eliminated due to high curvature of the road around the middle of the image (Figure 2d). 
It can be seen that the line images at different levels of the image hierarchy give different degrees of details of roads. 
Grouping of lines started from the top of the hierarchy, i.e., the image shown in Figure 2d, and the grouping results are 
  
  
Figure 2 (continued) Hierarchical grouping of line images e, f, g. grouping results at three different levels h. final road 
network after removal of non-road features. 
  
International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B3. Amsterdam 2000. 947