POM. 0 
Ce c 
C 
QO Lv € 
Xiunguang Zhou 
  
geometry distortion, an algorithm was proposed by the authors (X.Zhou and E.Dorror, 1994) to include orientation and 
scale in the cross-correlation algorithm. That is a time consuming algorithm. 
Though the feature based matching is free of geometric distortion, only a small number of points in images are belong 
to features. The matched points in the image are quite sparse. Attempts were made to combine the feature based and 
area based matching. In signal and feature combined matching algorithm, both gray values and features are used (Otto 
and Chau, 1989; Zong, Li and Schenk, 1992). The algorithm that proposed in this paper is also such a kind of matching 
algorithm that makes use of feature and area based matching together. 
A general problem in generating a DEM is that the obtained DEM loses a lot of topological details. This is because 
there are not enough matched points. Too many DEM points were actually obtained by interpolation using far 
neighbors. In order to get very detailed DEM, this algorithm tries to match every pixel in the image pair to provide a set 
of very detailed matched points for the DEM generation. In the following section, a general description of the algorithm 
is presented. 
2 GENERAL DESCRIPTION OF THE MATCHING ALGORITHM 
  
    
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
y eme, d 
Pyramid Generation Guidance To 
Based on WT Lower Levels 
zx cii 
Pyramids Lower ee n 
l Of Pyramids 
[] m] Geometry Differences 
Highest Level Determination Based on 
Of Pyramids Higher Level Matching 
i Results 
Edge Extraction Geometry Constraint 
And Description Cross Correlation 
Edge Matching N 
Geometry Differences Y 
Determination Based on 
Edge Matching Results Least Square Matching 
Geometry Constraint End 
Cross Correlation 
  
  
  
  
  
Fig. 1 Diagram of the algorithm 
The goal of image stereo is to measure the coordinate disparities of the pixels in the stereo image pair (the left image 
and the right image). It needs to determine the conjugate point pairs in the image pair. The two points in a conjugate 
pair must represent the same ground point. Usually, due to the different perspective views and different characteristics, 
  
International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B3. Amsterdam 2000. 1055