Fig. 2 Topographic map and Landsat-5 TM image of 
study area 
the 
4. 2.2 Definition of road-likeness 
As one of measure for representing road- 
likeness, Hough transformation (Ref. 6) was used. 
First spectral feature road is determined and then 
road segments were extracted. After that line- 
likeness and directionality were checked for the 
road segments followed by Hough transformation. The 
extracted line segments through Hough transformation 
were expanded by 3*3 pixels result in a 
representation of road-1ikeness(Ref. 7). 
4. 3 The results 
The small portion of image showing the location 
of the pure pixels of road, MIXELs and the other 
pixels are shown in Fig. 4(a) while the extracted 
road-like pixels were also shown in Fig. 4(b). The 
result from the determination of anchor points was 
shown in Fig. 5. Typical mixing ratios for the 
designated pixels which are indicated in Fig.6 are 
shown in Table 1 and show a reasonable proportion 
for the pixels in the paddy field which apart from 
the road, the pixels beside the road, the pixels in 
the road and the pixels at the disconnected 
■ 
1 
■ 
1 
■ 
■ 
_ 
r r\ a rl 
jffl i V P I 
(a) Preliminaly extracted pure road pixels, MIXELs 
and the other pixels 
portions. These information on proportion of the 
MIXELs are useful for contextual classification. As 
the matter of fact the disconnected portions are 
covered with the crown of the trees beside the road 
so that proportion of vegetation(paddy field) are 
relatively large compared to the pixels in the road. 
road 
(b) Extracted road, road-like and the other pixels 
Fig. 4 Extracted road, road-like and the other pixels 
345