The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B3b. Beijing 2008 
» is 30 
skewness value 
01« 
0 14 
0 12 
0» 
00» 
oce 
004 
oee 
probability 
•cropfield A 
3. Choose a threshold for binarization of the smoothed 
skewness map, and the higher valued part is the alternative 
region for urbanization. 
4. Combine the alternative region from skewness and 
multi-spectral bands, the urbanized area was extracted by 
ETM+ image. The old city region has a lot of vegetation, so the 
skewness feature is instable where the spectral feature is stable; 
through many tests, the new urbanized area has a more fixed 
skewness value, and it can be easily mixed with surrounding 
crop field. So, it was recommended that pure multi-spectral 
information is suitable for extract old urbanized area, while the 
new urbanized area need both skewness and multi-spectral 
information. 
Figure 4(b) Skewness distribution of crop field in Figure 3b and 
urban area in Figure 3e 
Figure 4(c) Skewness distribution of crop field in Figure 3c and 
urban area in Figure 3f 
4. ALGORITHM AND EXPERIMENT 
5. Use DEM data to exclude the area improper for construction 
from the alternative area, generated from the step 4. Since the 
bare rock in the mountains have very similar texture and 
spectral feature with new urbanized area, it was necessary to use 
slope data to delineate the boundary of the mountain area, 
where very few buildings can be constructed. 
6. Take morphological operation on the “urbanized area” 
generated from the step 5, and a reasonable urbanized area is 
determined at last. Thousands of scattered “urbanized patches”, 
in very small area, were distributed in the result. Such area 
might represent some targets of similar features with the “real 
urbanized area”, or they were actually urbanized area. To 
simply overcome this dilemma, erosion-dilation operator was 
implemented, since if a building is surrounded by large open 
space, it can not be regarded as “urbanized area” according to 
laws of urban planning. In the same way, while some other 
landcover, with small area, are surrounded by urbanized area, 
such landcover can be regarded as “urbanized area”, so a 
dilation-erosion operator is used to achieve this goal. Finally, a 
nearly perfect mapping of urbanized area was finished (Figure 
6). 
4.1 Technical Process 
Although skewness can effectively extract urbanized area, the 
misclassification also existed obviously. For instance, the 
skewness value of beaches around the Miyun Reservoir and 
some rocks distributed in mountains was similar to that of 
urbanized area. To correct this kind of errors, multi-spectral 
information from ETM+ and DEM data was used to eliminate 
such unurban area. The whole process is stated as follows: 
1. Select the proper window size for skewness computation on 
band 8 of ETM+. Because the band 8 (panchromatic) has a fine 
spatial resolution which contains abundant detail information, 
so it is the best band for extracting urbanized area. And the only 
parameter of the moving window can be controlled is the size, 
because a too small size can not reflect the symmetrical 
information, and if it is too large it will confuse different targets. 
Through a series of experiment, 9*9 size window was best for 
classification and computation. And then the skewness map was 
generated as Figure 5. 
2. Select a proper low pass filter to equalize the skewness value 
in patches. Since urbanized areas were distributed in patches, 
low pass operator can reduce the intra-distance of the same 
landcover. Unfortunately, the operator can also reduce the 
inter-distance of adjacent landcovers, so it is important to 
choose proper low pass filter for a trade-off. 
Figure 5. skewness map of ETM+ in Beijing 
465