The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B3b. Beijing 2008 
589 
Figure 8: Cross pixel structure 
3.2 Road Segmentation 
In automatic road extraction, each road section should be 
defined separately in a way that belonging pixels are grouped 
together representing a unique section. This task can be fulfilled 
in two stages: 
- Isolation of different road sections 
- Grouping road pixels into their relevant road sections 
3.2.1 Isolation of Different Road Sections 
Road section isolation can be performed in cross pixels where 
two or more road sections intersect. Considering the fact that 
road thinning has reduced the road width into one pixel, those 
pixels with more than three 8-neighbouring road pixels could be 
regarded as cross pixels (Figure 8). 
When cross pixels are found, they are removed to separate road 
sections into isolated segments. 
3.2.2 Grouping Pixels into Their Relevant Road Sections 
After road network separation, it would be possible to group 
pixels into their relevant pixels. It could be performed by 
sweeping the whole image and using run length encoding 
scheme. In the second sweep, neighboring runs are connected to 
each other to form a unique run and a run code is assigned. The 
strategy in run code assignment is to use successive positive 
integer numbers for road runs where non-road pixels are 
represented by zero values. . When neighboring runs are 
connected, coding is modified so that the maximum code is 
equal to the number of available isolated segments. Figure 9 
represent the coded run segments in different colors. 
Figure 9: Isolated run codes presented in different colors 
3.3 Automatic Vectorization Using CAD Environment 
Facilities 
In this research, vectorization of segmented road pixels was 
performed with the aid of MicroStation™ software. 
In general, vectorization process of segmented road sections in 
this software can be expressed as: 
- Coordinate assignment to all pixels belonging to each 
road segment. 
- Coordinate transformation to MicroStation™ software 
(to establish a relationship between road raster map 
and MicroStation™ environment). 
- Connecting successive points so that the segmented 
road would be represented as a multi line. 
Since road pixels belonging to each segment were categorized 
and coded in the previous section, each segment could be 
transformed independently into the MicroStation™. This task 
was performed by a designed software port to establish the 
relationship between programming and MicroStation™ 
environments. This port provides the system with all the 
facilities available in the MicroStation™ for vector-based data 
editing and data control processes. 
In digitizing a sample road segment, the first step is to find its 
beginning point. For this reason, the road raster map is swept 
top-down and left-right to find the first pixel belonging to the 
interest road segment. Then the pixels located in its direct 8- 
neighbourhood are followed in both directions until the starting 
point, having only one 8-neighbour, is found. Line following 
procedure begins from this point and road pixels are 
transformed into the MicroStation as™ separated points. 
Consecutive points are connected to each other to form a 
structured line string for each road segment. This vectorized 
line string contains all the geometric information of the 
represented road segment. This procedure is repeated for all 
road segments available in the road raster map. Figure 10 shows 
the vectorized road segments in the MicroStation™ 
environment.