ON MATCHING IMAGE PATCHES UNDER VARIOUS GEOMETRICAL CONSTRAINTS 
Charles K. Toth 
Toni Schenk 
Department of Geodetic Science and Surveying 
The Ohio State University, Columbus, Ohio 43210-1247 
USA 
Commission III 
ABSTRACT 
Image matching plays an important role in digital photogrammetry. Finding conjugate points occurs in different photogram- 
metric tasks. Image matching is usually performed in two steps: determining approximations and computing precise conjugate 
locations. In this paper we are concerned with the second aspect, that is, the image patches are already close to their final 
position. An image patch analyzer determines which matching primitives should be used first. Based on the results other 
primitives can be activated. The process terminates if a predefined level of confidence is reached or if no further improvements 
are to be expected. The matching process can be geometrically constrained, for example, along vertical lines, epipolar lines 
or by fixing one image patch in its location. 
KEY WORDS: Matching, Conjugate Points, Geometric Constraints. 
1. INTRODUCTION 
Image matching - finding conjugate points — plays an im- 
portant role in digital photogrammetry. It is an essential 
operation of many basic photogrammetric procedures, like 
automatic orientation or DEM generation (Schenk et al., 
1990). Much research in digital photogrammetry has been 
devoted to matching, including theoretical as well as im- 
plementation issues. The results are reported in numerous 
publications and technical papers. 
One of the first products of digital photogrammetry are dig- 
ital photogrammetric workstations (or softcopy stations). 
They will have a great impact on how daily photogrammet- 
ric tasks are handled (Kaiser, 1991). For one, they provide 
the operator with all functionality of the analytical plot- 
ters. The major difference between softcopy stations and 
analytical plotters is the fact that the operator views the 
3-D stereo model directly on the display screen. More im- 
portant, softcopy stations offer an unprecedented opportu- 
nity for automation. This is the first time that digital pho- 
togrammetric methods are implemented a in real production 
environment. This is as encouraging as it is challenging for 
the research community to transfer and commercialize re- 
search results. The process of automating routine tasks has 
just started. Since it is a very complex problem, results will 
probably not meet the high expectations for quite some time 
(Schenk and Toth, 1992). 
In this paper we report about the preliminary phase of a 
project with the objective to automatically keep the float- 
ing mark of a softcopy station on the ground (or dot on 
the ground, for short, DOG). The idea is that after a stereo 
model is oriented the operator is not forced to set the mea- 
suring mark (3-D cursor) precisely on the ground. If the 
function is evoked the system will do it automatically. To 
automatically place the floating mark on the surface is a 
problem of finding conjugate points in both image patches. 
In other words, when an operator slightly moves the floating 
mark in the XY plane, then the proposed algorithm should 
automatically find the corresponding Z coordinate and move 
the measuring mark accordingly (adjust Z). Consequently, 
these application-specific conditions immediately define the 
scope of the matching techniques feasible to solve the corre- 
spondence problem. We investigated the concurrent appli- 
cation of different matching methods with good localization 
results. In the following, conceptual issues and some major 
subtasks are discussed. 
2. OVERVIEW OF THE PROPOSED 
ALGORITHM 
2.1 Application Specific Conditions 
A variety of different matching methods are now available, 
each with its specific advantages and disadvantages. In order 
to narrow down the set of possible techniques suitable for our 
project, the application characteristics must be considered: 
e digital stereo model is oriented 
e good approximations for conjugate points are given 
e optional epipolar image geometry 
e parameter adaptability 
e relatively small size of the image patches 
The exterior orientation is necessary to move between image 
400 
and object space. It is used to constrain the movement of 
the floating mark. Also, it allows for employing object space 
matching methods (Helava, 1988). 
Since the operator will keep the measuring mark fairly close 
to the ground, good approximations for the conjugate points 
can be assumed. Thus, image patches always sufficiently 
overlap. 
As shown in (Kaiser, 1991) epipolar geometry can be easily 
achieved on softcopy stations, thus it is worth taking advan- 
tage of that special geometrical condition. Sometimes the 
image 
displa 
move: 
age p 
simila 
mean: 
ing) 1 
proce: 
surfac 
patch 
can re 
The x 
tional 
since : 
our a] 
cessin, 
would 
3.2.8 
Based 
match 
the cu 
shorte 
artifac 
tions : 
first k 
teristi