International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B3. Istanbul 2004 Intern 
Segments Filtering. To deal with big parking lots and buildings 
which are misclassified as road networks, a directional texture 
detector is developed to distinguish different types of objects 
according to their textures in different directions. The 
directional texture detector measures the pixel grey value 
variance along the central lines in each of four directions of an 
operation window. If all the variances in four directions are 
smaller than a certain value, it can be concluded that the object 
within this window is homogeneous. Therefore, this object can 
be considered as a non-road object and can be removed. Figure 
7 shows the input image and the processed result. The classified 
road network (Figure 7a) is the input. The result of the segment 
filtering is shown in Figure 7b. After segments filtering and 
edge-aided segmentation, the final road network is extracted 
(Figure 7c). 
  
(c) 
Figure 7. Directional texture detection and segment filtering. 
(a) The classified road network from the pan-sharpened image. 
(b) The classified road network after directional texture 
detection and segment filtering. (c) The extracted road network. 
3. TEST DATA AND RESULTS 
The images used in this study are QuickBird MS and Pan 
images taken in August 2002. The study area is Fredericton, 
NB, Canada. An example of the pan-sharpened image in the 
area is shown in Figure 8a with 1500x750 pixels. 
The classified road network (Figure 8b) is a mixture of roads, 
buildings, and grounds with many driveways connecting to the 
road network. They are difficult to remove using existing 
techniques. However, the proposed edge-aided classification 
demonstrates the ability to handle these problems. Figure 8c 
shows the final road network extracted using the proposed 
method. Almost all roads in the network are successfully 
extracted. Only two small parts of roads in the bottom left and 
upper right parts of the image are missing, due to insufficient 
road width. 
  
cla: