The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B3b. Beijing 2008 
471 
a Original image b Linear lifting wavelet c ASWlet lifting wavelet 
Figure 5. Different results of image decomposition using two kinds of lifting wavelet 
From figure 5b, we can see that there is very little non-zero 
information of high frequency portion in the decomposed image 
with linear lifting wavelet, which is useful to image 
compression. But it is difficult to show the features. Flowever, 
figure 5c shows more clearly the image features in the 
horizontal, vertical and diagonal direction through image 
decomposition with the ASWlet lifting wavelet. 
4.2 Image feature matching of ASWlet lifting wavelet 
parameters (Chen, 2000). 
3) Representing the edge feature using line moment and 
matching the image via parameters to get the initial value. 
4) Proceeding image reconstruction to prepare for the 
image matching on the next layer. 
5) Matching the image in next layer, which is similar to 2). 
6) Matching the image In the last layer using least square 
method and get the result in sub-pixel accuracy (Chen, 2006). 
The strategy of feature matching could be summarized as: 
1) Constructing the pyramid images using the 
above-mentioned ASWlet lifting wavelet for stereo image 
decomposition, and preparing for the layer matching. 
2) Computing the gradient map and direction map, 
extracting the feature and analyzing the vector, and then using 
the multi-scale analysis to restrain noise and histogram filtering 
to get the feature map which is used for feature matching; On 
5. BUILDING MODEL AUTOMATIC EXTRACTION ON 
THE STEREO IMAGES 
With the multi-scale edge detection based on the wavelet 
analysis, comer point recognition and feature matching, we 
studied the method of automatic extraction of the 3D building 
geometry model. Figure 6 shows one flow chart of 
semi-automatic building extraction on the stereo images 
Figure 6. Flow chart of semi-automatic extracting building on the stereo images 
From a stereo image, via above-mentioned handling, we can intersection. At last we perform the building geometrical 
obtain a series of corresponding image points of key points on a outline extraction, 
building (such as comer points and inflection points). The 3D 
coordinates of these points could be obtained by stereo