Hee Ju Park 
  
    
— 
This definition has empirical character. If the point pair relation between other point pairs is similar, 8M, will be 
relatively high. If the relation is weak then SM J will be small compared to other point pairs. 
Point pairs whose SM is lower than a threshold are eliminated, for example the pairs with lowest 5% in SM value. 
33 Line matching 
If we can reduce the number of candidate lines to be compared, we reduce the processing time and increase the 
possibility of finding corresponding line pairs. In line matching, firstly we reduce the number of possible candidate lines 
as described in the following: If the relation between any matched point and a pair of lines on both image is reversed 
like Figure 5, this case of line can be discarded. One line has two flanking regions see Figure 3. If a pair of lines is 
corresponding, then at least one flanking region of them is similar in each colour channel. So line pairs without any 
similar flanking regions can be rejected . 
Because we can know the positions of many overlapping windows by matched point sets in previous stage, we know 
the approximate difference in each RGB colour channel value for whole corresponding regions using the matched 
points. We represent the differences as mean and standard deviation for each colour channel. We assume that the 
following case rarely happens for each colour channel. 
chine IdGij— E(dG)l» 4* sdv(dG) (15) 
o 
dGij: flank region difference between line i and line j 
ur fix E(dG)» sdv(dG) : Mean & standard deviation of corresponding region difference 
ting of 
et al If both flanking regions of a line pair fulfill the above case, then we can discard such a line pair in the comparison. We 
atched consider the line pairs which pass the test as two check points. : 
If the In the next step we calculate the similarity coefficient for the overlapping part between the lines for the local similarity 
hod is comparison. There are many possible definitions for this similarity coefficient. 
points One simple definition is as follows. Suppose the coordinates of a point on the line of image i as (xX, y,), and its 
better ; ; ; : : : 
elation corresponding point as (X, yj) . Then y, — y; without any rotation except through relief displacement in stereo 
r each epipolar imagery. The line similarity coefficient C; can be defined as 
ear the du (16) 
_ COV(X;,X;) 
sij cij nij 
S, Ss x; 
W.; weight for the number of overlapping flanks. (e.g. W., 21 when one of the line flanks is similar. And 
global W,=15 when both flanking regions are similar for every colour channel.) 
M ef W, : weight for the number of overlapping points between line i & j, i.e. the number which has same y 
ear the coordinate each other between line i & j. 
Jud) | ean ki | 
This above definition doesn't consider the effect due to slope in object space. This slope may be straight or curve. To 
p J p p y g 
simplify comparison, we suppose average slope in height. Suppose there is angle c due to average slope like Figure 6. 
The bigger average slope in height is, the bigger the angle q is. 
(14) 
line i line j 
left flank left flank 
right flank right flank 
Image 1 Image 2 Image 1 Image 2 
Figure 3. Flanking region of a line Figure 4. Neighbouring points and lines in line matching 
  
International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B3. Amsterdam 2000. 701