ISPRS, Vol.34, Part 2W2, '‘Dynamic and Multi-Dimensional GIS", Bangkok, May 23-25, 2001 
377 
4(a) Original SPOT Stained by Noise 
The figure 3(a) is a SPOT panchromatic image acquired from 
Wuhan. There are obvious objects such as roads and some fields 
in the image. The figure 4(a) results from figure 3(a) stained by 
random noise. The figure 3(b), 3(c) and 3(d) are the results which 
are respectively processed by the classic Sobel, Robert 
operators and the a trous wavelet decomposition method. In the 
condition of noise, we use the methods and get the results shown 
in figure 4(b), 4(c) and 4(d). Comparing with the figure 3, we can 
find the main edges of the figure 3(d), for example, the two roads 
are finer than those in the figure 3(b) and 3(c). The tiny field 
edges are obvious in the figure 3(d), but they do not exist in the 
figure 3(b) and figure 3(c). From the results of figure 3, the edges 
detected by the proposed method are finer and it can detect the 
tiny edges. 
From the figure 4(b) and 4(c), we find the random noise effect is 
apparent, but almost no the noise effect in the figure 4(d). 
Therefore the classic Soble and Robert methods can not filter 
noise, they are sensitive to noise. The & trous wavelet 
decomposition method can detect edges and filter random noise. 
The major reason may be that the big features of the image 
almost don’t verify at the different scales of wavelet transform, but 
the random noise rapidly attenuates with the increasing scales; 
on the other hand, the random noise is weakened by adding 
some wavelet planes. Therefore, selecting proper scales, we can 
detect the edges and overcome random noise. In some 
applications we have used the results detected by the £ trous 
wavelet decomposition method and the results are satisfactory. 
3(b) Sobel Operator 
3(c) Robert Operator 
3(d) A trouse wavelet decomposition 
Fig3 SPOT Image and Detected results 
3(a) Original SPOT Image