AUTOMATIC DEM DATA GENERATION AND CHANGE DETECTION 
BY REGION MATCHING 
K.C.LO , N.J. MULDER 
LT.C. 
P.O.Box 6, 7500 AA Enschede 
The Netherlands 
(ISPRS Commission III) 
ABSTRACT: 
The automatic DEM generation from stereo image pairs by computer aided matching is a fast and economic 
method. But the matching would fail when it is performed within the homogeneous intensity region, or in 
the regions which the land cover has changed when the satellite stereo pairs are taken with a long interval 
of time. This research tries to use the Conditional Rankorder Operator to smooth the intensity within the 
region first, then start Region Growing for Image Segmentation, the boundary of region can be extracted, 
and the shape can be described by y-s Curve, combine with other properties of region, such as the area, the 
position of gravity centre, etc., to form a Property List. The initial probability of region matching can be 
obtained by Minimum Cost Function with the weighting properties in the list. Then the matching 
probabilities are being adjusted by Relaxation Processes until the final conjugated region pairs are 
determined. The elevation of the region can be calculated with the conjugated region, and the Change 
Detection can be done by checking the mismatching regions and by comparing the intensity between the 
conjugated region after region matching. 
Key Words : Region Matching, Segmentation, Region Growing, w-s Curve, Relaxation Process. 
As a result of the inherent linear differential operator 
1. INTRODUCTION involved, a common limitation of linear operators is 
their amplification of high spatial frequency noise and 
The automatic DEM generation from digital stereo artifacts. This situation can be improved by applying a 
images by stereo matching is a fast and economic low pass filter averaging mask based on regions of pixels 
approach. But the low level intensity matching would fail first, but this might lead to smoothen the line/edge as 
when it is performed within a homogeneous intensity well as the noise. 
region, or in the regions in which the land cover has The alternative method can be that smoothing the 
changed when the satellite stereo pairs are taken with a relative homogenous area and then enhancing the 
long interval of time. Therefore, the boundary of the edge/line. This would prevent the disadvantages previ- 
homogeneous intensity region should be detected and ously mentioned. We can select the statistical (e.g. take 
extracted for high level feature matching. After region mean) characteristics with non-linear operators (i.e. 
matching, the disparity between the conjugated region nonlinear combinations of pixels). For example, the 
can be obtained for DEM generation. In the meantime, Average Smoothing is a method by ‘local clustering 
we confirm the regions of mismatching after region [Sijmons, 1986]. It is a straight forward way to remove 
matching or by checking the intensity within the conju- noisy pixels and smooth a region by giving each point a 
gated region, the change of land cover can be detected. new grey value which is the average of the original grey 
value in some neighbourhood of a point (e.g. 8 neigh- 
bourhood points) and with the point itself being included. 
2. IMAGE SMOOTHING/ENHANCEMENT This is a powerful method to smooth the region, remove 
5 ; the noise and detect the edge which is the boundary 
For helping the region segmentation, we need to reduce where the lower grey value abruptly changes to a high 
noise and the small variation of grey value within the grey value (or vice versa). It will blur the thick lines, and 
region on the one hand and enhance the edge features on suppress thin lines as well as isolated points. It also 
the other hand. We review the linear operator, such as "clips" corners, after successive applications of the 
differential operators (e.g. Laplacian operator). There operator [Lo, 1985]. Because we like to retain the distinct 
are some disadvantages, namely : linear features which are good for matching purposes, we 
L Blurring of the nearly homogenous background with / can modify this method from linear into non linear 
without noise interference. smoothing by adding a condition and call it the Condi- 
2. The grey value and characteristics of linear features tional Smoothing method. This method includes the 
have been changed, it would be influenced by its back- condition that if a line/edge occurs in the sub-image 
ground texture if we do not adjust the kernel of the during the convolution of the image, and the degree of 
operator according to the different backgrounds. distinction (i.e. the grey value difference between the 
476 
neig 
thre 
the 
Bec: 
whic 
slow 
a b 
adja 
tion 
to tl 
[Mu 
Bas 
imp 
enh: 
the 
subi 
valu 
with 
the 
Gen 
do f 
get : 
sort 
repe 
unti 
sele 
Mec 
proj 
(thi 
still 
valu 
esti1 
tion 
met 
ord: 
bou: 
[SD 
shri 
hoo 
of if 
pror 
opel 
rem 
(the 
purr 
orig