The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B7. Beijing 2008 
1192 
Figure 5. ERS-2 SAR image of Shanghai, China 
Figure 6. Landsat ETM+531 image of Shanghai, China 
An ERS-2 SAR data (5.3GHz, VV polarization, spatial 
resolution 12.5m) on the Aprial 9,2002 and a Landsat ETM+ 
data (spatial resolution is 30m) on the November 27, 2002 over 
the Central Park of Shanghai Pudong District, China are taken 
as an example. The ERS-2 SAR image is shown in Figure 5 and 
the combination of band 1, 3 and 5 of Landsat ETM+ image is 
shown in Figure 6. It can be seen that it is difficult to 
distinguish the road from linear water at the top right of the 
microwave radar image. 
The crossroads in the upper and lower left of Figure 6 present 
that some roads are broken by the shadows of the vegetation 
and house surrounding them in the multi-spectral image. 
Besides, the fact that there are some other objects whose 
spectral features are similar to the road also makes it difficult to 
extract the road from only multi-spectral image. 
Figure 7. The result of the iteration difference for Landsat 
ETM + 531 image 
Figure 7 presents the spectral feature extracted from the multi- 
spectral image with the iterative difference method of this paper. 
It shows that the value of the road and house in the lower left of 
the image is larger than other objects and makes them easy to 
be extracted from the images. 
Figure 8. Texture features extracted from Landsat ETM +image 
Figure 8 shows the texture extracted from the entropy of the 
gray-level co-occurrence matrix of the multi-spectral image. 
The house with big and flat surface presents smaller value while 
the road edge and the region with abundant texture present 
larger values. 
Figure 9. Distribution of spatial size for ERS-2 SAR 
Figure 9 is the spatial auto-correlation scale of ERS-2 SAR 
image calculated with multi-layer and multi-scale Getis spatial 
autocorrelation method. The red or yellow color in the image 
shows the positive spatial auto-correlation scale, resulting from 
the cluster of the buildings. Green or blue color shows negative 
spatial auto-correlation, resulting from the cluster of such 
objects as grassland, flat surface, wide road or river with 
smaller backscattering characteristics. 
Figure 10. Fusion image of Landsat ETM + and ERS-2 SAR