and 
'tric 
'tric 
PRS 
C). 
lg à 
Ahmed Elaksher 
INTEGRATION OF OBJECT AND FEATURE MATCHING FOR OBJECT SURFACE 
EXTRACTION 
Pakorn APAPHANT', James BETHEL'" 
National Research Council Thailand 
Remote Sensing Division 
Pakorn ? hammerhead.nrct.go.th 
"Purdue University 
School of Civil Engineering, USA 
Bethel @ecn.purdue.edu 
KEY WORDS: Matching, Digital Elevation Model, DEM, Dynamic Programming, Signal Matching, Feature 
Matching 
ABSTRACT 
A stereo matching algorithm is developed for object surface construction. The matching problems are addressed by the 
integration of signal and feature matching. The innovative strategy arises within the framework of global optimization 
of the match function. A conventional feature matching method based on dynamic programming is investigated and 
extended in this research. Several types of primitive features are extracted and matched. The object coordinates from the 
results of feature matching are used as weighted constraints during the signal matching process. In order to evaluate the 
performance of the algorithm, images of both urban and rural scenes have been tested. The experiments have shown 
promising results using this approach. 
1 INTRODUCTION 
The construction of 3D urban area spatial models from aerial images is a difficult problem that continues to challenge 
researchers in the image understanding and photogrammetry fields. The main problems are occlusion, large parallax 
ranges, and feature textures which are not amenable to the matching process. An urban area image by definition 
contains numerous cultural features. Feature based matching has been used to address these problems. However, if only 
matched feature information is exploited, it is not possible to reconstruct a surface model completely. Signal based 
matching has been used for solving such problems as well. 
In order to achieve results in the presence of significant occlusions, the feature matching problem can be addressed via 
dynamic programming, DP. This technique permits orphan features which have no match on the other photograph. The 
preliminary object surface generation can then be reduced to the problem of an optimal profile search from a cost matrix 
generated by features from a given stereo pair. This matching technique generally consists of three parts which are (1) 
extracting features to be matched along with their descriptive attributes, (2) generating a correspondence cost matrix, 
and (3) searching for the optimal path in the cost matrix. 
In this investigation, some of the problems found in stereo matching, as mentioned above, will be addressed by a 
strategy of combined signal and feature matching based on a global or semi-global search technique. The sequence of 
steps for the feature matching component would entail extracting features along epipolar lines in the images, and 
matching the features as suggested above by a DP-based approach. The feature coordinates in the object space are then 
determined. After that a signal matching technique with feature constraints is applied to generate the elevation profiles 
via another variant of dynamic programming. Once all profiles in a model are determined, the surface in the object 
space can finally be constructed. 
2 FEATURE EXTRACTION 
Several types of low level features are extracted from images, along with their positions in epipolar space. They are low 
level features which contain useful information for the matching process. Some features used in this research are 
already well defined from elsewhere, such as the straight line feature, and some are developed here such as end points 
of a plateau feature, and the spike point feature. For all of these features, either the extraction is done directly in the 1D 
epipolar space, or the extraction is done in the 2D image space, and subsequently intersected with the 1D epipolar lines. 
  
International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B3. Amsterdam 2000. 267